The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
u -E o c o U LII 1&1 FEDERAL FEDERAL RESERVE BANK BANK OF OFDALLAS DALLAS THIRD THIRD QUARTER QUARTER 1995 1995 Measuring the Policy Polley Effects EHects Of Changes in In Reserve Requirement Ratios joseph jo.wph H. J/. Hastag Ilaslag and anti Scott SroII E.E. Hein Jlein ANew AlIew Quarterly Output Measure for Texas TellS Franklin Fr(ll/illitl D. D. Berger and grid Keith IfJ!ilb R. Phillips Alternative Methods Of Corporate Control Commercial Banks In Commerclaillattks Stephen D. Prowse SltfJben D. ProtI'St! This publication was digitized and made available by the Federal Reserve Bank of Dallas' Historical Library (FedHistory@dal.frb.org) Economic Review Federal Reserve Bank 01 Dallas Robert D. MeTllr, Jr. Pr!ll(lenl ¥III CfrieI toe:uMl ()jig Toay J. IImlDIo FI!'II Vb PrmldenI n1 CNII ~ IlIIta' Hlmy Rosenilam SonIer VQPr8!lll!nlIIIII~ It ~ W. MIalIlICU: Ykz PrIsDrt RI Ean:mcAlt/1501 SllpMo P. A. Bnrn AssiIIIIII: YIaa PmicIi!o'.I ¥Id Senior E~ RnlltCfl Oft'lcen John Ouca Robert W. GitmeJ William C. Gruben Evan F. Koenig Ecenomlltl Kenneth M. EIII81Y Bevertv J. FOK David M. Gould Joseph H. Haslag O'Aoo M. Pelefsen Keith R. Phillips Stephen O. Prowse Fiona D. Sigalla Lori L Taylo! lucinda Vargas Mafk A. Wynne Mine K V!k:el carfos E. Zaralllgil - Raurdl Aasaclma Pr~ NaItw1 S. I!aIke ..... ..... ................ PrDlessor Thomas B. Fomby SoIAIen~u.~ Prolessor Gr800IY W. Hultman Professor Fioo E KVdland ~o! T_ilAuslin Prolessor Roy J. Rullin IWYs'sIIy II! HwsIM Editors Stephen P.A. Brown Evan f . Koonig Monaglng Edllor Rhonda Harris Copy Edhor Monica Reeves Grapldc OnIgJl -"'" laJra J. Bell The Ecoooo!ic ~ is plOiiftd by !he FodtraI Resero'8 &nk 01 Dalla$. The I'iews ~ are !hose 01 the authors af'I:I do 00I1leIZS5¥'1y rdIed Ite posi. tioos 01 IIle feclefal fIe:ser,oe Bank 01 Dallas or ItIe fedef~1 Rewve System, Subscriplioos .t il'/allable Irte d charge. Please send requests tor sinole-capy a~ multiple.copy Sl.tIwIjllions, batk issues, and a1Ihss changes 10 IhI! Publi~ Allalr$ Oep.Jrtmert, Federal Reserve Bank 01 03l1as. P.O. 8oJ. 655905. Dallas, TX 752f6.-!i!ni. (214) 922-52!i7 Articles may be ~Inted 00 lhe condition Ilia! Ihe soma is crecilleci and Ihe Re5eiH;h ~ar1men1 is provided with a cOjIjI oIlhe publicalioo contain ing the reprinted material Contents Measuring the Policy Effects of Changes in Reserve Requirement Ratios Joseph H. Haslag and ScOIl E. Hein Page 2 A New Quarterly Output Measure for Texas Franklin D. Berger and Keith R. Phillips Page 16 Alternative Methods of Corporate Control in Commercial Banks Stephen O. Prowse Page 24 The monetary hase is (he sum of high-powered money and an adjustllll:n! factor that measure.s changes in reSt:lve requirement ratios. This adjustment factor is calculated .so lhm it responds 10 changes in deposit k:vcls in addition 10 changes in reserve requirements. Cons(;!quently, researchers and policym;tkcrs using the monel:l ry base are seeing;t mixture of changes impkmentcd through open market operations, uiSCoLlnt window borrowings, and reserve fequiremcntli, together with nonpolicy actions acting on deposit flows. Joseph Hasbg and Scott J-1cin calcul<ue the reserve step index (RSD to s<:par:He changes in one of the availahlc adjustment factor::; the 5t Louis Federal Heserve Hank's Reserve Adjustment Me:lsurc O~AM )-int o pure rescrve-requirt!ment effects and deposit-now effects. HSI would give analysts ;1 measure (hat responds only to chang!.!.. in reserve requiremem ratios. Haslag and Hein also provide statistical evidence s Ll ~esling Ihal combining RSI and the depositnow effecl , :IS RAM docs, is not justifiablt! in simple reduced-form mooeb of nominal GNP growth, otl1put growth , or innation. Heal gross ciOlllt:slic product is one of the most watched indiC;Hor.; of the U. S. husineS$ cycle Yet at the stale level. output mea$U~S are rardy USt-d to track husiness conditions. Although the Bureau of Economic Analysis estimates real gros.<; sta tt! product (RGSP), the long n:lease lag (usually about two and one-h;llf years :Ifler the reporting ye:lr) and the annual periodicity of the data severely limil its u$efulness. In this ,Iltide, Frank Berger and Keith Phillips find that movements in quarterly personal income and various price me:lsures ca n accurately explain movements in total Texas RGSP and in eleven broad industry groupings. Based on these findings, Berger and Phillips create quarterly measures oftoral and industry-specific Texas RGSP that will be ,lv,libble about four months ,Ifter rbe report ing quarter The new series represents :, comprehensive me:lSUfC of economic activity in the .<;tate that can be used along with olherlimely indicators, such :IS nonfarm employmcflI and (he unemployment rate, to gauge current business conditions. In Ihis article, Stephen Prowse investigates how owners of commercial hanks enCOllf<lge management to follow value-maximizing policies While the "corporate control mechanism" in nonfinancial Finns is well documented , fo r the banking industry much less evidence is available. Moreover, unique faclOrs in the operating e nvironment ofcormnen:ial ba nks may mean that their corporate control mechanism oper.Hes differently from that of no nfin:mcial fi rms. Prowse analyzes a sa mple of hank holding companies (BHCs) from 19H7 to 1992 to detennine how many underwent a change in corporate conlrol by hostile takeover, friendly merger. action by the board of dire<.tors. Of interyention by regulators. Prowse finds that the prim:lry market-based corpor.lfe control mt."Chanism among BJ-ICs is action by the board, although hank boarels appear to be much less assertive than hoards of nonfinancial firms. Over-Ill, the market-based corporate control mechanism.~ in banks do not appear as effident at disciplining manager... as they are in other firms . By default, Ihis has given a primary role to regulators to provide a "1:tst resort" control mechOlnism. Prowse analyzes ~asons for this and ev:t!ua tes bmv proposed hanking leJ.iislation mighl affect corpomte governlmce. Karl Brunner (1961) argues that movements in a monetary policy measure should solely reflect those actions undertaken through all three of the Federal Reserve’s policy tools: open market operations, discount window loans, and changes in reserve requirements. In Brunner’s view, highpowered money—the sum of bank reserves and currency held by the nonbank public (also known as source base)—is too narrow a measure for policy analysis. His main criticism is that changes in reserve requirement ratios would not cause movements in high-powered money. As a result, he suggests constructing an adjustment factor— which he terms liberated reserves —to measure policy actions undertaken via changes in reserve requirements. Brunner defines the monetary base as the sum of high-powered money and the adjustment factor. This combination provides a monetary policy measure that possesses Brunner’s desired property of representing all Federal Reserve tools.1 Following Brunner’s lead, both the Federal Reserve Bank of St. Louis (hereafter “St. Louis”) and the Board of Governors of the Federal Reserve System (hereafter “Board”) currently calculate monetary base series that add an adjustment factor to high-powered money. The purpose of the adjustment factor ostensibly is to measure, in dollar terms, monetary policy actions implemented through changes in reserve requirements. In both the St. Louis and Board measures, the adjustment factor is an index value constructed as the difference between what required reserves would have been under the base-period reserve requirement structure and actual required reserves. Movements in the index value, therefore, are interpreted as changes in the amount of required reserves freed (absorbed) relative to the base period. Peter Frost (1977) and Manfred Neumann (1983), however, have argued that the St. Louis index value is a poor proxy for measuring changes in reserve requirements.2 In essence, these critics argue that movements in the adjustment factor, over time, can occur for nonpolicy reasons. Critics claim, as such, that the adjustment factor is not a pure measure of policy changes conducted through the Federal Reserve’s tools but includes other considerations. This measurement issue is potentially important for students of monetary policy. If movements in the adjustment factor are an amalgam of policy and nonpolicy actions, it is a mistake to interpret movements in the St. Louis adjustment factor as a direct measure of reserve requirement ratio changes. To illustrate this point, suppose that a nonpolicy action causes a movement in the Measuring the Policy Effects of Changes in Reserve Requirement Ratios Joseph H. Haslag Senior Economist Federal Reserve Bank of Dallas Scott E. Hein Norwest Bank Distinguished Scholar Texas Tech University T his measurement issue is potentially important for students of monetary policy. If movements in the adjustment factor are an amalgam of policy and nonpolicy actions, it is a mistake to interpret movements in the St. Louis adjustment factor as a direct measure of reserve requirement ratio changes. 2 adjustment factor. A researcher looking at such an episode in monetary policy history would, using the adjustment factor, erroneously identify this movement as reflective of a policy action. This identification problem also arises when researchers estimate correlations between the adjustment factor and economic variables, claiming the adjustment factor measures reserve requirement ratio changes. If the adjustment factor does not properly distinguish between policy and nonpolicy actions, it is not clear whether either episodic differences or the estimated correlations are due to movements in policy actions, nonpolicy actions, or both. Examples of potential inference problems arise in several articles examining the relationship between reserve requirements and economic activity, including Prakesh Lougani and Mark Rush (forthcoming), Joseph Haslag and Scott Hein (1992), Charles Plosser (1990), and Mark Toma (1988). We have two main objectives in this article. First, we describe and construct a measure of changes in required reserves caused by changes in reserve requirement ratios for the United States from 1929 through 1993. Unlike the current reserve adjustment measures, this alternative measure distinguishes between movements resulting from changes in reserve requirement ratios and those resulting from changes in deposits. Our alternative measure is constructed by modifying Brunner’s liberated reserves notion. More specifically, we constrain changes in the reserve index measure to equal zero during periods in which no changes in reserve requirement structures were implemented. Our objective is to generate a cleaner measure of changes in reserve requirements, especially for analysts explicitly interested in monetary policy research. The second, and more important, objective is to empirically assess the importance of this measurement issue. While the criticism of the existing procedure has been around since the late 1970s, the significance of the distortion has been generally ignored. After providing a descriptive (episodic) overview of the measurement differences, we use formal statistical techniques to quantify the differences between the existing adjustment factor and our measure. Brunner (1961) first suggested the idea of a comprehensive measure of monetary policy actions. He proposed the notion of liberated reserves, defined as reserves freed or impounded by changes in reserve requirements. Leonall Anderson and Jerry Jordan (1968, 8) describe the process of constructing the reserve adjustment measure for a particular month as follows: First, the weighted average reserve requirement on demand deposits for the month (using for weights the distribution of these deposits by class of member bank) is computed. Then, the difference in average reserve requirements from the previous month is multiplied by net demand deposits for the previous month. The reserve adjustment measure is then the algebraic sum of the monthly estimations. Thus, the change in the reserve adjustment measure is simply (1) where L denotes liberated reserves, Dt is the period-t level of deposits against which reserves are required to be held, and ∆wr = wrt – wrt –1 is the change in the weighted average of reserve requirement ratios. 3 (The ∆ is the first-difference operator.) The weighted average takes into account that reserve requirements are different for different-sized banks. For example, in 1994 the first $51.9 million of checkable deposits at a particular bank are subject to a 3-percent reserve requirement. (This $51.9 million level is called the low-reserve tranche.) For deposit levels above the low-reserve tranche, the reserve requirement ratio is 10 percent. The weighted average is then the sum of the following products: the fraction of period-t deposits that are subject to the lowtranche reserve requirement times 0.03 (the 3percent reserve requirement) and the fraction of period-t deposit levels that are above the lowreserve tranche level times 0.10 (the 10-percent reserve requirement). Note that if there were only one reserve requirement ratio, this approach would yield changes in liberated reserves only when reserve requirements were changed. A problem with constructing liberated reserves in this way is that nonpolicy actions can affect the change in liberated reserves over time. For example, suppose that depositors shift their accounts from banks with deposit levels below the low-reserve tranche to large banks. Because the reserve requirement ratio is higher at the large The history of adjustment factors Before we describe the alternative method used to construct the reserve requirement change, it is useful to provide a brief overview of the adjustment factor. With such an overview, one can better understand how the definition of adjustment factor has evolved over time and the criticisms of this measure. FEDERAL RESERVE BANK OF DALLAS ∆L = Dt ∆wr, 3 ECONOMIC REVIEW THIRD QUARTER 1995 bank than at the small bank, the weighted average reserve requirement ratio will change. Consequently, there is a change in the reserve adjustment measure. Frost (1977) identifies this problem, as well as another concern, in measuring monetary policy using the monetary base measure. (See the box entitled “Frost’s Logarithmic Adjustment Factor” for details on his methodology.) At about the same time as Frost’s work, two researchers at the St. Louis Fed, Albert Burger and Robert Rasche (1977), were calculating the Reserve Adjustment Magnitude, or RAM. St. Louis’ RAM was designed to measure the impact of changes in reserve requirement ratios. The Federal Reserve Bank of St. Louis’ adjustment measure is calculated as (2) In the remainder of this article, we turn our attention to measuring changes in required reserve ratios. We construct our measurement by constraining the change in measure of required reserves to equal zero for those periods in which no change is made in reserve requirements. Constructing the reserve step index We offer an alternative measure of effective reserve requirement ratios that is not affected by such deposit flows. To create our measure, which we term the reserve step index (RSI), we use data on weekly levels of required reserves. We modify the St. Louis base-period selection process; that is, our RSI measure is constructed using the reserve requirement ratio structure for August 1978. St. Louis, however, uses the average reserve requirement structure for 1976 –80 as its base period.6 We choose August 1978 for our RSI base period for two reasons. First, August 1978 is the month in which RAM is closest to zero (there are no monthly values of RAM in which it is identically zero). Second, our base-period selection permits a fairly direct comparison with the St. Louis RAM measure insofar as the average reserve requirement ratio in 1976 –80 is evidently close to the August 1978 reserve requirement structure. The various dates for changes in reserve requirement ratios are obtained from the annual report of the Board of Governors for every year from 1929 to the present.7 With the dates of the changes in reserve requirement ratios, the difference between required reserves in the week(s) in which the change in structure took place and the week prior to change is used as our estimate of the value of reserves freed (absorbed) by the policy action.8 This measure is added to the previous level of RSI, resulting in a cumulative measure of dollar changes in required reserves resulting from changes in reserve requirement ratios. We consider separately the dates after August 1978 and the dates before August 1978. For dates after August 1978, we look for the first period that reserve requirement ratios were altered. For each date after August 1978, we calculate the change in RSI as RAMt = (rb – rt )Dt , where rt is the vector of the reserve requirement ratios that apply in period t, rb is the vector of the reserve requirement ratios in a selected base period, and Dt is the vector of period-t quantity of deposits against which reserves must be held. The St. Louis monetary base adds RAM to highpowered money. RAM can be interpreted as the level of required reserves in period t less what required reserves would have been were the base-period reserve requirement still in effect. A positive value of RAM indicates that reserves have been freed relative to the base-period reserve requirements. Conversely, a negative value indicates that reserves have been impounded relative to the base-period reserve requirements. Consider how RAM changes over time. The changes in RAM from one period to the next (discrete time) can be represented as (3) ∆RAMt = –∆rt Dt + (rb – rt )∆Dt . Equation 3 indicates that RAM changes over time in response to two factors. The first term on the right-hand side of equation 3 captures changes in reserve requirement ratios. The second term, which we refer to as the deposit-flow effect, indicates that RAM can change over time even if reserve requirements are constant.4 Specifically, changes in deposits indirectly reflect both the households’ and banks’ behavior. Because these changes affect both RAM and the monetary base, the adjustment factor presents a basic identification problem: movements in RAM can be due to changes in reserve requirements, due to changes in deposits against which reserves must be held, or some combination of both. The implication is that RAM is a potentially poor proxy of changes in reserve requirement ratios.5 (4) RSIt – RSIt –1 = RRt –1 – RRt if reserve requirement changes, or 0, otherwise. Similarly, from the August 1978 benchmark, we move backward in time, looking sequentially for dates on which changes in the reserve requirement ratio structure were implemented. Again, equation 4 is used to determine the path of RSI. 4 Frost’s Logarithmic Adjustment Factor Peter Frost (1977) identifies a problem with a monetary base measure. Specifically, Frost argues that the measure “distorts the effect of Federal Reserve policy actions on the growth in the money supply” (1977, 168). Frost’s point is that the growth rate of the monetary base should be equal to the growth rate of high-powered money in periods in which reserve requirements are constant. Yet, Frost shows that differentiating the log of liberated reserves with respect to time yields (A.1) money. Formally, (A.3) Differentiate equation A.3 with respect to time and solve for the growth rate of the logarithmic adjustment factor (L*) to yield (A.4) Γt = t ∑ Gjr ∆rt , j =1 where Gr is the arithmetic mean of 1/(r + k ) for period t and period t –1, r is the reserve-to-deposit ratio, k is the currency-to-deposit ratio, and ∆r is the change in reserve requirements. Obviously, with ∆r = 0 (a case in which reserve requirements are constant over time), Γ is constant. The logarithmic reserve adjustment measure uses Γ so that the percentage change in the logarithmic reserve adjustment is equal to the change in high-powered In periods in which there is no change in reserve requirements, equation 4 dictates that the step index be held constant. When the change in RSI is not zero, equation 4 indicates the change in the dollar amount of reserves freed (absorbed) by changes in reserve requirements in the particular week in which the reserve requirement change was enacted. While the input data we use is weekly, our aim is to construct a monthly series. We simply sum across all the weekly changes in RSI that take place within a month to get a monthly value. With the estimates of monthly changes, we start with RSI = 0 in August 1978, adding the monthly value of the change in RSI to the previous month’s level, both forward and backward in time, to create our time series. The result is an index time series documenting the cumulative measure of FEDERAL RESERVE BANK OF DALLAS dL* L *dB dΓ = + (B + L*) . dt Bdt dt From equation A.2, the term d Γ/dt = 0 during periods in which reserve requirements do not change. Thus, the percentage change in L* is equal to the percentage change in B, implying that B and B + L* grow at the same rate. For our purposes, equation A.4 indicates that the value of L* does change over time, even when reserve requirements do not. So, Frost’s approach satisfies the criterion that growth rates for high-powered money and monetary base are identical in those periods in which reserve requirements do not change. However, L* is not a good indicator of changes in reserve requirements, in our sense, because it moves over time even though reserve requirements do not change. It is important to note that Frost’s Γ term is quite similar to our notion of what makes a good measure. Both RSI and Γ do not change during periods in which reserve requirement ratios are constant. In constructing RSI, we use the level of required reserves as the basis for calculating changes in required reserves caused by changes in reserve requirement ratios. Implicitly, we are multiplying changes in reserve requirement ratios by deposits. However, Frost calculates Γ as the product of the change in reserve requirement ratios and 1/(r + k ). ∆B ∆B ≠ for L ≠ 0. B+L B Frost proposes a solution to this problem: a logarithmic adjustment factor. The bottom line is that Frost’s series ensures that the monetary base and high-powered money grow at identical rates in those periods in which reserve requirement ratios do not change. However, in levels, the logarithmic adjustment factor moves over time, even in those periods in which reserve requirements do not change. As such, Frost’s measure is suspect as a measure of monetary policy actions implemented through changes in reserve requirement ratios. Frost’s adjustment factor is defined as (A.2) ln(B + L*)t = lnBt + Γt . changes in required reserves, relative to August 1978, that are due to changes in reserve requirement ratios. The relationship between changes in RSI and changes in RAM is straightforward. First, note that RRt = rt Dt (where RR is required reserves). For periods in which changes in reserve requirement ratios occur, substituting this expression into equation 3, one can write (5) ∆RAMt = rb’∆Dt + ∆RSIt for rt ≠ rt –1, where ∆ is the difference operator. In periods in which no changes in reserve requirements take place, rt = rt –1, (6) ∆RAMt = rb’∆Dt – rt’∆Dt = (rb – rt )∆Dt , and ∆RSIt = 0. 5 ECONOMIC REVIEW THIRD QUARTER 1995 Taken together, equations 5 and 6 describe the movements in RAM, differentiating between periods in which changes in reserve requirement ratios occur and periods in which only changes in deposit levels occur. Equations 5 and 6 share a common term, rb’∆Dt . This term represents the change in required reserves due to deposit flows (∆Dt ). The deposit-flow effect, (rb – rt )’∆Dt , makes it difficult to use the St. Louis measure as proxy for measuring changes in reserve requirement ratios. Using the RAM methodology, required reserves are treated as freed (or absorbed) as deposits change over time, even when reserve requirement ratios are fixed and rb ≠ rt . In contrast, for periods in which no changes to reserve requirements occur, ∆RSI = 0 by construction.9 Our next objective is to empirically assess the costs, if any, of including the deposit-flow variable in a measure of changes in reserve requirement ratios. Because both series are index numbers, a comparison of the two series is implicitly a comparison relative to the base period. Because we use essentially the same base period to construct RSI as RAM, however, absolute comparisons of the two series, and the implied reserve requirement ratio structures, are a justifiable approximation. The two reserve index series (RAM and RSI) exhibit qualitatively similar time series behavior. Some important differences, however, emerge during particular episodes. For example, consider the period 1929–36. This period represents an interval in which there is a sizable difference between the levels of the two measures. RSI hovers around –$4 billion for most of this period, indicating that reserve requirements were higher during this interval than those in place during the 1978 base period. In contrast, in the 1929–36 period, RAM is near zero for the entire period. A researcher using RAM (and interpreting it as changes in reserve requirement ratios) would infer that reserve requirements in the 1929 through 1936 period were really not that different from those of the 1976 –80 period. The interpretation provided by RAM is that reserve requirements were not very restrictive during the first half of the 1930s relative to the August 1978 base period. RSI, however, suggests a much more restrictive policy stance was in place in the 1929–36 period relative to the August 1978 base period. Specifically, the 1929–36 reserve requirement structure absorbed about $4 billion Comparing RSI and RAM over time Figure 1 plots the original RAM series and RSI, the step index, from January 1929 through December 1993. (The actual monthly series for RSI is included in the appendix.) By construction, RSI does not move in periods between changes in reserve requirement structures. As such, the reserve step index is constructed as a sequence of infrequent, permanent shocks. This time series behavior is quite different from that of RAM, which shows much more drift; that is, RAM experiences more frequent changes in its level. Figure 1 RAM and RSI Series, January 1929 – December 1993 Billions of dollars 40 35 30 RAM RSI 25 20 15 10 5 0 –5 –10 ’29 ’31 ’33 ’35 ’37 ’39 ’41 ’43 ’45 ’47 ’49 ’51 ’53 ’55 ’57 ’59 ’61 ’63 ’65 ’67 ’69 ’71 ’73 ’75 ’77 ’79 ’81 ’83 ’85 ’87 ’89 ’91 ’93 SOURCES: Federal Reserve Bank of St. Louis and authors’ calculations. 6 in reserves relative to the August 1978 reserve requirement structure. After 1936, both RSI and RAM decline, and by 1945, the two series obtain about the same level. In highlighting the 1929–36 period, it is easy to see inference problems created by the presence of the deposit-flow effect during the Great Depression. The fact that RAM is close to zero during the 1929–36 period has more to do with the outflow of deposits from banks during this period than with monetary policy actions. As such, one could wrongly infer that reserve requirements were about the same in the 1929–36 period as they were during the 1976 –80 period.10 Another discrepancy between the time series behavior of RAM and RSI occurs beginning in 1973 and ending about 1975. In early 1973, RSI falls slightly, while RAM begins a steady decline that ends in early 1975. Both RAM and RSI are below zero, indicating that the reserve requirement structure during the 1973–75 period was high relative to the appropriate base periods. Deposit inflows, with basically high reserve requirements relative to the base period, drive RAM down sharply from 1973 through 1975. RSI, however, indicates that very few required reserves were absorbed by reserve requirement ratio changes during this period. As measured by RAM, the rapid deposit growth in this period would have exaggerated the policy constraining effects of reserve requirements. As the public moved deposits into reservable deposit accounts in 1973, RAM suggests that the level of required reserves was becoming more and more restrictive during the 1973–75 period. RSI suggests, however, that the average level of required reserves was raised only slightly between 1973 and 1975.11 Similarly, again in 1987–90, deposit outflows drive RAM sharply lower than RSI. Between 1987 and 1990, RAM increases only slightly. One could infer that reserve requirement ratios had been lowered slightly, freeing a small amount of reserves. In contrast, RSI rises rather sharply during the 1987–90 period, indicating that monetary policy was actually freeing a larger amount of reserves.12 Based on RAM, the late 1980s looks like a period in which reserve requirements were lowered slightly, then held fairly steady. In contrast, RSI indicates that a series of policy actions was implemented in which reserve requirements were lowered. The data in Figure 1 can be reconciled by treating the small increase in RAM as resulting from deposit outflows. Reserves freed by lower reserve requirements were being offset by smaller quantities of deposits. Consequently, RAM—the product of these two sepa- FEDERAL RESERVE BANK OF DALLAS rate effects —shows only slight increases, while RSI accurately captures the falling reserve requirements. A time series analysis of the differences between RSI and RAM By displaying the levels of the RAM and the RSI series, Figure 1 depicts episodic differences between the two measures. However, the evidence does little to shed light on the importance of such differences. It may be the case that the two series only randomly deviate from (a linear combination of ) one another. One way to shed light on this issue is to examine the statistical long-run relationships between RSI and RAM. In particular, we ask whether RAM and RSI are similar time series. Our belief is that there is no permanent long-run association between these two measures in the sense that if one wanted to forecast movements in RSI using RAM over an infinite horizon, the variance around that forecast would be infinity. Based on the time series behavior presented in Figure 1, we suspect that deposit-flow effects can, and do, cause the two variables to permanently diverge from one another. Robert Engle and Clive Granger (1987) have suggested the use of cointegration techniques to explore long-run relationships in time series. If deposit-flow effects are a short-run phenomenon that results only in temporary deviations between the two measures, then deviations between the two series should disappear in the long run. Hence, such deviations can be characterized as simply “noise.” On the other hand, if deposit flows are significant and not self-reversing, there is likely to be no long-run relationship in the two series. Evidence indicates that both RAM and RSI are integrated of order one—I(1).13 As such, the two series may be cointegrated. The following is the output from an ordinary least squares regression using levels of RAM and RSI (standard errors in parentheses): (7) RAMt = –.323 + .862 RSIt + et . (.125) (.17) D –W = .03; R 2 = .82. As the two measures, RSI and RAM, each has unit roots, a test for cointegration seeks to determine whether there is a unit root in the residual, et , from equation 7. Under the null hypothesis that there is a unit root in et , the test statistic is –2.25, which is larger than the 5-percent critical value of –3.17. Hence, one fails to reject the null hypothesis that there is a unit root in the error 7 ECONOMIC REVIEW THIRD QUARTER 1995 term. Thus, one cannot reject the null hypothesis that RAM and RSI are not cointegrated.14 Similarly, the small value of the Durbin–Watson statistic also suggests that RAM and RSI are not cointegrated. The evidence, therefore, suggests there is no long-run relationship between RAM and RSI. As such, there is no evidence to support the notion that RSI and RAM are driven by a common factor. Nor should one conclude that deviations in the two measures simply reflect selfreversing noise. Our interpretation of these tests is that RAM gives weight to a deposit-flow effect that may permanently bias the estimate of changes in required reserves resulting from true changes in reserve requirement ratios over an infinite horizon. Moreover, the evidence suggests that the deposit-flow effect is itself integrated of order one; that is, the deposits against which reserves must be held have a unit root, resulting in RAM and RSI not being cointegrated. Indeed, an auxiliary test is to look for unit roots in the differences in the time series —RAMt – RSIt — which, by construction, is the deposit-flow effect, measured as the vector product of deposits against which reserves must be held times the difference in reserve requirements in the current period and in the base period. Unit root tests on this variable indicate that this deposit-flow measure is indeed integrated of order one —I(1). In light of these findings, the discrepancy in the two alternative measures of changes in required reserves is not a trivial issue. The differences between the two series do not gravitate toward zero in the long run. The evidence presented formalizes what “ocular econometrics” suggests —the two series are different. This evidence further suggests that the two measures would provide very different signals about changes in reserve requirement ratios over time. cal. The question, however, is whether this constraint is empirically supported. To examine the first question, we begin by separating the deposit-flow effect and the reserverequirement effect. We define ∆DEPFLOWt = ∆RAMt – ∆RSIt . ∆DEPFLOW generally will not equal zero for periods in which changes in reserve requirement ratios occur.16 The strategy here is to estimate reducedform macroeconomic models in which the explanatory variable, the percentage change in RAM, is decomposed into the percentage change in the deposit-flow variable and the percentage change in RSI (each as a proportion of the adjusted monetary base).17 The reduced-form setting is useful for purposes of identifying differences in predictive content. In these simple regressions, we are focusing on the indicator properties of the separate components of the monetary base. (Of course, this question does not answer whether the monetary base is a better or worse indicator compared with other variables.) We estimate separately reduced-form models of the inflation rate (using the implicit price deflator), the percentage change in real GNP, and the percentage change in nominal GNP. The right-hand-side variables in these regressions are lagged values of the percentage change in high-powered money (∆SB), ∆RSI, and the deposit-flow variable (∆DEPFLOW ). In the inflation and output growth equations, we include both lagged values of the inflation rate and real GNP growth. In the nominal GNP growth equation, lagged values of nominal GNP growth are also included. We use the Akaike Information Criterion to select the appropriate lag length for all explanatory variables in the regressions. The general representation of the reduced-form regressions is as follows: n1 n2 n3 j =1 j =1 j =1 (8) ∆Yt = a0 + ∑ ∆Yt − j + ∑ ∆SBt − j + ∑ ∆RSI t − j Relationships to economic activity In this section, we examine two specific questions in a reduced-form macroeconomic setting. First, does the deposit-flow measure help to predict movements in macroeconomic variables differently from the current reserve requirement measure?15 Since RSI ignores the deposit-flow effect, testing for marginal predictive power of deposit-flow effects is an indirect test of what measure is contributing to the predictive power of RAM. Second, are the coefficients on depositflow measure equal to the coefficients on RSI? Because RAM essentially constrains these two effects to be equal, empirical work using RAM supposes that the effects of changes in reserve requirements and changes in deposits are identi- + n4 ∑ ∆DEPFLOW j =1 (9) t−j , n1 n2 n3 j =1 j =1 j =1 ∆Pt = a0 + ∑ ∆Pt − j + ∑ ∆y t − j + ∑ ∆SBt − j + n5 n4 ∑ ∆RSI j =1 t−j + ∑ ∆DEPFLOWt − j , j =1 and n1 n2 n3 j =1 j =1 j =1 (10) ∆y t = a0 + ∑ ∆Pt − j + ∑ ∆y t − j + ∑ ∆SBt − j + ∑ ∆RSI j =1 8 n5 n4 t−j + ∑ ∆DEPFLOWt − j , j =1 Table 1 Regression Results for Inflation, Output Growth, And Nominal GNP Growth Equations, 1929–83 where ∆Y is nominal GNP growth, ∆P is the inflation rate, and ∆y is output growth. The n i ’s denote the appropriate lag length for each variable in the model. Unfortunately, national income and product accounts data are not constructed in a consistent manner back to 1929. Hence, we use two different data sources. For the period 1929–83, we use quarterly data from Nathan Balke and Robert Gordon (1986) on real GNP and the fixedweight deflator. Since these data end in 1983, we also consider a postwar period that includes more recent history, namely, 1951–93. For this period, we use real GNP, the implicit price deflator, and nominal GNP data from the Bureau of Economic Analysis. In all, we estimate the three reduced-form regressions over two periods: 1929–83 and 1951–93. The key reason for separating the reserverequirement effect and the deposit-flow effect in a reduced-form specification is that the depositflow effect signals changes affecting both the demand for deposits by households and businesses and the supply of deposits by banks. As such, the deposit-flow effect in this reduced-form equation is not a pure policy measure but is an amalgam of these different shocks.18 The reduced-form setting used in this analysis does little to shed light on the transmission mechanism differentiating the reserve-requirement effect from the deposit-flow effect because we do not have structural equations. However, the reserve-requirement effect represents a tax on the banking system and, hence, is a particular type of shock. Thus, while these tests do not provide direct evidence on the structural effects, they do provide evidence on whether separating reserve-requirement effects from other effects helps to predict economic activity. Moreover, the ∆RAM measure implicitly assumes that the effects of changes in ∆DEPFLOW and ∆RSI are equal. By separately including the deposit-flow measure and ∆RSI in the regression, we can test whether this restriction is supported by the data. This test is important in analyzing the response of macroeconomic variables to policy changes, as given in impulse response functions. Specifically, if the coefficients on lagged values of ∆RSI are different from the coefficients on lagged values of the deposit-flow measure, one cost of using ∆RAM is that impulse response functions — or, for that matter, any parameter estimate—will be biased. Table 1 reports the sum of the estimated coefficients for each of the three regression equations, using data for the period 1929–83.19 The table also summarizes evidence on the null FEDERAL RESERVE BANK OF DALLAS Sum of the Estimated Coefficients Dependent Variable Independent Variable Inflation Inflation Real GNP .005(1) NA –.051(2) .515(3) NA ∆SB .091(6) .005(3) .172(6) ∆RSI .176(7)** .560(2)** .568(7)** ∆DEPFLOW .062(1) Output growth Nominal GNP growth .720(1)** Nominal GNP –.409(1)* NA NA .160(2) .627(4)** Table 2 Regression Results for Inflation, Output Growth, And Nominal GNP Growth Equations, 1951–93 Sum of the Estimated Coefficients Dependent Variable Independent Variable Inflation Real GNP Nominal GNP Inflation .867(3)** Output growth .064(1) .310(1) NA –.001(1) .007(1) .005(1) .008(7) –.010(7) –.030(7) –.037(7) .002(1) –.006(1) NA .366(1) ∆SB ∆RSI ∆DEPFLOW Nominal GNP growth –.135(1) NA NA * Significant at the 10-percent level. ** Significant at the 5-percent level. NA denotes not applicable. NOTE: Numbers in parentheses represent the number of lagged values included in the regression. hypothesis that the sum of the coefficients equals zero. Insofar as the sum of the coefficients indicates some long-run relationship present in the reduced-form equation, the test determines whether there are significant long-run predictive effects.20 Table 1 documents that ∆RSI is significantly related to changes in inflation, output growth, and nominal GNP growth in the sense that the sum of coefficients is different from zero. Increases in ∆RSI, occurring because of a lowering of reserve requirements, predict subsequent increases in inflation, output growth, and nominal GNP growth. In contrast, ∆DEPFLOW is not related to the inflation rate or nominal GNP growth. Moreover, while ∆DEPFLOW is weakly related (at the 10-percent significance level) to output growth, as indicated by the tests on the sum of the coefficients, increases in deposit flows predict subsequent decreases in output growth.21 Table 2 reports the sum of the coefficients and test statistics, estimating the same relationships with data after World War II: 1951–93.22 9 ECONOMIC REVIEW THIRD QUARTER 1995 Table 3 Tests of Exclusion Restrictions Panel A: Sample Period 1929– 83 Dependent Variable Inflation Real GNP Nominal GNP 3.54** 6.96** 4.86** .62 2.76* 3.06** ∆RSI 4.26** 2.29** 2.63** ∆DEPFLOW 6.84** .49 .66 Independent Variable ∆RSI ∆DEPFLOW Panel B: Sample Period 1951– 93 * Significant at the 10-percent level. ** Significant at the 5-percent level. NOTE: F-statistics calculated under the null hypothesis that coefficients on lagged values of independent variable equal zero. Tests are conducted on the same regressions that are reported in Table 1 and Table 2. The results differ from the 1929–83 period in two particular ways. First, the sum of the coefficients on ∆RSI and ∆DEPFLOW is uniformly smaller in the 1951–93 sample than in the 1929–83 sample. Second, neither ∆RSI nor ∆DEPFLOW exhibits a statistically significant long-run relationship to economic activity in the 1951–93 sample. Another way to distinguish between reserve-requirement effects and deposit-flow effects is to determine whether they differ in terms of their short-run predictive content. Specifically, the question is whether movements in one or both of the components of RAM help to predict changes in economic activity. The test statistic, sometimes referred to as a Granger causality test, is calculated under the null hypothesis that the coefficients on lagged values of the variable are jointly equal to zero. The test indicates whether reserve requirements or deposit flow, or both, can help predict short-run changes in economic activity. Table 3 reports the test statistics for both our samples. In Panel A, the tests are reported using the 1929–83 sample, whereas Panel B reports the tests calculated using the 1951–93 sample. In both sample periods, the results indicate that ∆RSI always helps to predict changes in inflation, output growth, and nominal GNP growth. The ability of changes in ∆DEPFLOW to predict changes in economic activity is uneven across the two samples. Changes in ∆DEPFLOW help to predict nominal GNP growth and are marginally related to output growth in the 1929–83 sample, but they are not significantly related to inflation. In the 1951–93 sample, changes in ∆DEPFLOW are significantly related to changes in inflation but are statistically unrelated to movements in output growth and nominal GNP growth. Thus, the evidence suggests that either both reserverequirement effects and deposit-flow effects contribute to a relationship between RAM and economic activity, or only reserve-requirement effects contribute to RAM’s predictive content. These results suggest differences between reserverequirement effects and deposit-flow effects, but the evidence relates to predictive content and does not bear on whether the two effects should be separated. Presumably, one would want to distinguish between the two effects if combining the two into one measure throws out useful information. The next step is to directly test the hypothesis that changes in required reserves resulting from changes in reserve requirement ratios (as measured by ∆RSI ) have the same regression coefficients as those of the changes resulting from deposit flows. These results bear on the issue of whether there is a need to separate the reserverequirement and deposit-flow effects. One interpretation is that these coefficients describe the short-run dynamics when shocks hit the system. In vector autoregressions (VARs), these parameter estimates are used to generate impulse response functions. Thus, coefficient equality tests examine whether the short-run dynamic effects of changes in reserve requirement ratios should be constrained to equal the effects of changes in the deposit-flow variable. Table 4 reports F-statistics from two different tests. In the joint hypothesis tests, the test statistic is calculated under the null hypothesis 10 Table 4 Test Statistics on the Equality of RAM and DEPFLOW Coefficients that all individual coefficients on ∆RSI and ∆DEPFLOW are equal to one another. On the other hand, the sum of the coefficients test determines whether the sum of the coefficients on lagged values of ∆RSI is equal to the sum of the coefficients on lagged values of ∆DEPFLOW. As such, the first test examines whether significant differences in the short-run dynamics are present, while the second test examines whether the longrun impacts are statistically different for the two effects. Because the reduced-form models use stationary time series, it is unlikely that movements in ∆RSI and ∆DEPFLOW will result in long-run changes in inflation, real GNP growth, and nominal GNP growth, as stationarity implies that each series would return to its timeindependent mean values. In Table 4, the top half reports the tests for the 1929–83 sample, while the bottom half reports the findings obtained using the 1951–93 sample. In all six cases, the statistic for the joint hypothesis rejects the null hypothesis, suggesting that the coefficients on lagged values of ∆RSI are not equal to coefficients on lagged values of ∆DEPFLOW. The short-run predictive content of the two variables is very different. Only in the case of output growth in the 1929–83 sample would one reject the null hypothesis that the sum of the coefficients is equal. Thus, the evidence for output growth in these two samples rather strongly rejects the notion the short-run dynamic path following a shock in RSI is identical to the path following a shock to ∆DEPFLOW. On the other hand, the general evidence suggests that the longrun effects generally are not significantly different from one another. We interpret the significant differences between the coefficients on ∆RSI and ∆DEPFLOW as evidence against combining the reserverequirement effect and deposit-flow effect, as is done in RAM. Thus, a measure of reserve-requirement effects is useful for looking at pure policy effects. Estimating period: 1929– 83 Equation Sum of the Coefficients Inflation 3.51** .81 Output growth 8.17** 15.73** Nominal GNP growth 3.65** .81 Inflation 4.14** 1.10 Output growth 2.27** .36 Nominal GNP growth 2.34** 1.27 Estimating period: 1951– 93 ** Significant at the 5-percent level. 1 Reported is an F-statistic calculated with degrees of freedom (n, 213 – n) for the 1929– 83 sample and (n, 138 – n) for the 1951– 93 sample, respectively. Here, n is the number of restrictions placed on the regression. reserve requirements are not changing. The basic problem is that present approaches to quantifying these policy effects are influenced by depositflow shifts. In particular, decisions under the purview of the public or the banking community result in shifts among deposits with different reserve requirements that will result in changes in the current Federal Reserve System measures. Over time, the accumulation or decumulation of deposits changes the measures even though reserve requirements are unchanged. Why has this problem with the current measures been ignored? One can surmise that the economics profession either does not believe the criticism is valid or, alternatively, believes the measurement error is trivial. The purpose of this article is to challenge this conventional wisdom. To develop our case, we first construct our own reserve requirement step index (RSI), thus providing an alternative to the measures constructed by the Federal Reserve Bank of St. Louis and the Board of Governors. Our reserve requirement step index excludes, by construction, the most significant movements that result from deposit-flow occurrences. For purposes of comparison, we construct our index for the period 1929–93. We compare our measure with the conventional measure used by the Federal Reserve Bank of St. Louis. There are several distinct historical episodes in which significant differences between RSI and the St. Louis measure exist. The evidence suggests that significant deposit flows have had rather large impacts on the conventional measures over time. Moreover, we document that such measurement distortions are not temporary but are, indeed, quite long lasting. Summary and conclusions For many years now, monetary economists have recognized the value of quantifying the effects of reserve requirement ratio changes in measuring the monetary base. In fact, the Federal Reserve System currently provides such a measure to the public. Yet, the methodology used to quantify the effects of reserve requirement ratio changes has been criticized, dating back to at least 1977. Researchers have pointed out that current approaches to quantifying the effects of reserve requirement ratio changes are flawed to the extent that these measures can change even when FEDERAL RESERVE BANK OF DALLAS Joint Hypothesis1 11 ECONOMIC REVIEW THIRD QUARTER 1995 The differences in the two measures further result in different statistical associations between macroeconomic variables. We find that the purer measure of reserve-requirement effects generally has strong statistical associations with both inflation and real GNP growth. The deposit-flow effects, which cloud the measures of reserve requirement changes under the current methodologies, do not have a similar strong relationship to these macroeconomic variables. In fact, the evidence suggests that the statistical relationships are quite different for the reserve-requirement effects and the deposit-flow effects. As such, our findings raise serious concerns about using conventional measures of the monetary base, which presume the effects of reserve requirements and deposit flows are the same. The secondary aim of this article is to provide a better measure of reserve requirement ratio changes. Our efforts in this vein should be viewed as an approximation. One would need individual bank data to accurately measure the reserve-requirement effect. Based on our approximation, however, the conventional wisdom is challenged; that is, the measurement of reserve requirement ratio changes does not represent a mere second-order concern. Current approaches are not sufficient statistics for reserve requirement ratio changes. In contrast to the general view, we believe that it is useful to provide an accurate time series of reserve requirement ratio changes and that this measurement is potentially an important issue for economists in many macroeconomic, monetary, and financial applications. 3 4 5 6 Notes 1 2 The authors thank John Duca, Milton Friedman, Bill Gavin, Rik Hafer, Evan Koenig, Allan Meltzer, Manfred Neumann, Stephen Prowse, Dan Thornton, and Mark Wynne for helpful comments on earlier drafts of this article. Any remaining errors are our own. George Tolley (1957, 466) also discusses a measure of changes in the average reserve requirement ratio. His construction of average reserve requirements is “jointly determined by government, banks, and the non-bank public.” However, Tolley’s concept of an average reserve requirement ratio is quite different from Brunner’s. Tolley includes currency in his definition of reserve base. Thus, currency has a reserve requirement ratio of 1, and Tolley’s notion is more like a money multiplier, though he refers to it as a reserve requirement ratio. Brunner focuses on liberated reserves exclusively through changes in reserves that are required against deposits, rather than both currency and deposits. This criticism applies equally to the Board adjustment factor. Because essentially the same criticisms apply, we focus on the St. Louis measure. Moreover, Haslag 7 8 12 and Hein (1992, forthcoming) provide evidence suggesting that the St. Louis measure is more closely related to macroeconomic activity, in a statistical sense, than the Board measure. Here, we take the liberty of treating the adjustment factor equations as if there is one kind of deposits against which reserves must be held. In reality, there are different types of deposits (for example, savings and demand deposits) and bank characteristics (for example, reserve city, nonreserve city, small) that determine the level of a particular bank’s required reserves. Anderson and Jordan are careful to specify that the construction procedure for demand deposits also applies to savings accounts. The appropriate vector representation of D and r are omitted without loss of insight into the problems we are identifying. Neumann (1983) shows that RAM suffers from the same problem that Frost identifies with liberated reserves; that is, the growth rate of the monetary base is not equal to the growth rate of high-powered money during those periods in which there are no changes to reserve requirement ratios. In addition, Neumann cites the dependence of measuring current monetary policy on the cumulated sum of past changes in reserve requirements. This also is an interesting measurement problem, but our current focus is solely on the deposit-flow issue. The Board of Governors uses the current reserve requirement ratio structure as the base period. In an earlier article (Haslag and Hein 1990), we provide evidence suggesting that the St. Louis approach is more closely related to nominal GNP growth than the Board measure. Our belief is that the St. Louis base has a closer statistical relationship to nominal GNP growth than the Board base because the 1976–80 base period is more representative of the average reserve requirement ratio structure than today’s structure. We gratefully acknowledge the help of Dennis Mehegen at the Federal Reserve Bank of St. Louis for providing us with weekly data for the period January 1968 to June 1991. In going through the Board of Governors’ annual reports, we have selected all changes in reserve requirement structure, including definitional changes and size changes. See Joshua Feinman (1993) for a partial list in which the major reserve requirement changes are identified. As such, our reserve step index is still subject to deposit-flow effects, but only to the extent that deposit flows occur between the weeks in which reserve requirement ratios are changed. Note that in equation 2, the RSI can be rewritten as rt –1Dt –1 – rt Dt . Changes in deposits from t– 1 to t will be picked up in the RSI. To correct for this deficiency, one would need the use of detailed deposit data that are not generally available. Specifically, one would need data on deposit levels by type for each bank. This is necessary because differ- 9 10 11 12 13 14 ent reserve requirements have applied to different deposit levels. We believe that this deposit-flow effect is small, especially relative to the deposit-flow effects present in current reserve adjustment indexes that permit change in months in which no reserve requirement ratio changes occur. It should be noted that RSI represents essentially an average, as opposed to a marginal, concept. If one were to divide RSI by the quantity of deposits against which reserves must be held, the term would represent the average reserve requirement ratio. To construct a marginal reserve requirement series, detailed data are necessary on the quantity of deposits held at individual banks by each reserve requirement distinction. Such data, however, are not available. Thus, our efforts yield a first approximation of changes in average marginal reserve requirement ratios. Interestingly, Milton Friedman and Anna Schwartz (1963, 526) characterize the 1936 reserve requirement hikes as significant factors in the slowdown in economic activity that began in 1937, an interpretation that is more consistent with the behavior of RSI than RAM. Several changes in reserve requirement structure were enacted during the 1973–75 period. In short, reserve requirements on demand deposits were raised slightly in 1973, lowered in 1974, and lowered in 1975. On balance, reserve requirements were lowered for the smallest deposit levels ($0 to $2 million) and for the largest banks (over $400 million), with the intermediatesized deposit levels experiencing no change in reserve requirements. Haslag and Hein (1989) suggest that the Monetary Control Act of 1980 (MCA) effectively lowered the average reserve requirement for all depository institutions. Our RSI measure supports the inference that MCA effectively freed reserves for the system. In other words, each series is differenced once and is stationary—I(1). The Phillips–Perron test is applied to RAM and to RSI in both level and percent-change forms to examine the order of integration of each series. RSI is designed as a series in which there are infrequent, permanent shocks. Asymptotically, the distribution theory behind the unit root tests applies to series such as RSI. However, Nathan Balke and Thomas Fomby (1991) argue that standard Dickey– Fuller critical values result in too many rejections of the unit root null hypothesis in finite samples. Under the null hypothesis that there is a unit root in RAM, the test statistics are 0.80 in level form and –25.67 in percent-change form, whereas the test statistics are 1.88 in level form and – 46.38 in percentchange form for RSI. The 5-percent critical value is –3.17. The evidence suggests that RAM and RSI are nonstationary in levels but stationary in percent change. When RSI was regressed on RAM, the evidence similarly failed to reject the null hypothesis that the series were not cointegrated. FEDERAL RESERVE BANK OF DALLAS 15 The issue is intertwined with differences between outside and inside money. The deposits against which reserves must be held are liabilities of banks and thus reflect changes in the demand for and supply of intermediated deposits. Changes in reserve requirement ratios, other things held constant, affect the demand for high-powered money. Robert King and Charles Plosser (1984) distinguish between real and monetary effects, arguing that changes in outside money (the monetary base) are nominal changes, whereas movements in inside money (the money multiplier) represent real changes in the financial intermediation process. King and Plosser find that the monetary base is correlated with prices but not with real economic variables. However, the money multiplier is closely correlated with real economic variables. Scott Freeman and Greg Huffman (1991) provide a theoretical model that yields the same qualitative correlations as King and Plosser find. In this case, both changes in reserve requirements and changes in deposits against which reserves must be held are real changes. However, one is a policy variable, and the other may only reflect behavioral changes due to policy changes. 16 The changes in required reserves during the week in which reserve requirements are changed will generally not equal the difference between RAM measured during the month in which reserve requirements changed and the month before the change occurred. The calculation of percentage change relative to the quantity of monetary base is as follows: ∆SBt = (SBt – SBt –1)/[(MBt + MBt –1)/2], where SB denotes high-powered money and MB denotes the monetary base. Thus, ∆MBt = ∆SBt + ∆RSIt + ∆DEPFLOWt . Note that the variables are stationary in percentchange form. In a structural setting, Eugene Fama (1982) argues that a bank’s decisions to supply deposits is likely to be related to changes in reserve requirements. By having both deposits and reserve requirements in the regression, we are implicitly examining the effects of changes in reserve requirements on economic activity separately from the effects of changes in deposits. Table 1 attempts to provide some idea of the regression results without going into too much detail. Reporting the sum of the coefficient saves space compared with reporting each individual coefficient. The full set of parameter estimates and the data series are available from the authors upon request. The regressions are run with variables that are stationary. With stationary series, the thought experiment seems a bit odd. The tests on the sum of the coefficients determine whether a once-and-for-all movement in the policy would be related to a permanent change in the measure of macroeconomic activity. Yet, the policy variables have not exhibited movements that are 17 18 19 20 13 ECONOMIC REVIEW THIRD QUARTER 1995 21 22 consistent with the sort of permanence suggested by the thought experiment. Indeed, both the policy measures and macroeconomic variables have reverted to their constant mean value. An important issue is the stability of the regression coefficients over the sample period. We follow the approach taken by Martin Feldstein and James Stock (1993). We use a battery of six different tests for parameter stability. Further, we treat the exact date(s) at which the parameters changed as unknown. In each of the three models estimated (inflation, output growth, and nominal GNP growth), the test statistics fail to reject the null hypothesis that the parameters are constant over the sample. The test statistics are available from the authors upon request. of Monetary Aggregate to Target Nominal GDP,” NBER Working Paper Series, no. 4304 (Cambridge, Mass.: National Bureau of Economic Research, March). Freeman, Scott, and Gregory Huffman (1991), “Inside Money, Output, and Causality,” International Economic Review 32 (August): 645–67. Friedman, Milton, and Anna J. Schwartz (1963), A Monetary History of the United States, 1867–1960 (Princeton, N.J.: Princeton University Press). Frost, Peter A. (1977), “Short-Run Fluctuations in the Money Multiplier and Monetary Control,” Journal of Money, Credit, and Banking 9 (May): 165–81. We adopt 1951 as the starting point because the Treasury– Fed accord establishes an identifiable change in the Federal Reserve’s operating procedure. Note that including data back to 1947, when quarterly data for the postwar period become available, does not change the major results presented in this article. Haslag, Joseph H., and Scott E. Hein (forthcoming), “Does It Matter How Monetary Policy Is Implemented?” Journal of Monetary Economics. ——— (1992), “Macroeconomic Activity and Monetary Policy Actions: Some Preliminary Evidence,” Journal of Money, Credit, and Banking 24 (November): 431–46. References Anderson, Leonall C., and Jerry Jordan (1968), “The Monetary Base: Explanation and Analytical Use,” Federal Reserve Bank of St. Louis Review, August, 7–14. ——— (1990), “Economic Activity and Two Monetary Base Measures,” The Review of Economics and Statistics 72 (November): 672–76. Balke, Nathan S., and Thomas B. Fomby (1991), “Infrequent Permanent Shocks and the Finite-Sample Performance of Unit Root Tests,” Economic Letters 36 (July): 269–74. ——— (1989), “Federal Reserve System Reserve Requirements, 1959–1988,” Journal of Money, Credit, and Banking 21 (August): 515–23. ———, and Robert J. Gordon (1986), “Appendix B: Historical Data,” in The American Business Cycle: Continuity and Change, ed. R. J. Gordon (Chicago: University of Chicago Press), A1–A6. King, Robert G., and Charles I. Plosser (1984), “Money, Credit, and Prices in a Real Business Cycle Model,” American Economic Review 74 (June): 363–80. Lougani, Prakesh, and Mark Rush (forthcoming), “The Effect of Changes in Reserve Requirements on Investment and GNP,” Journal of Money, Credit, and Banking. Brunner, Karl (1961), “A Schema for the Supply Theory of Money,” International Economic Review 2 (August): 79–109. Neumann, Manfred J. M. (1983), “The Indicator Properties of the St. Louis Monetary Base,” Journal of Monetary Economics 12 (August): 595–603. Burger, Albert, and Robert Rasche (1977), “Revision of the Monetary Base,” Federal Reserve Bank of St. Louis Review, July, 13–28. Plosser, Charles I. (1990), “Money and Business Cycles: A Real Business Cycle Interpretation,” in Monetary Policy on the 75th Anniversary of the Federal Reserve System, ed. Michael T. Belongia (Norwell, Mass.: Kluwer Academic Publishers Group), 245–74. Engle, Robert F., and Clive W. J. Granger (1987), “Cointegration and Error Correction: Representation, Estimation, and Testing,” Econometrica 55 (March): 251–76. Fama, Eugene F. (1982), “Inflation, Output, and Money,” Journal of Business 55 (April): 201– 31. Tolley, George S. (1957), “Providing for Growth of the Money Supply,” Journal of Political Economy 65 (December): 465–85. Feinman, Joshua (1993), “Reserve Requirements: History, Current Practices, and Potential Reform,” Federal Reserve Bulletin 79 (June): 569 –89. Toma, Mark (1988), “The Role of the Federal Reserve in Reserve Requirement Regulation,” The Cato Journal 7 (Winter): 701–18. Feldstein, Martin, and James H. Stock (1993), “The Use 14 Appendix Monthly RSI Series, January 1929 – December 1993 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 Jan. Feb. Mar. Apr. May June July Aug. Sept. Oct. Nov. Dec. –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.107 –8.955 –5.41 –5.41 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –5.115 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.787 –2.115 –2.115 –3.657 –3.331 –3.331 –1.107 –1.522 –1.296 –.274 –.17 0 –3.184 –3.963 –1.955 –.326 3.498 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 25.813 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.107 –8.955 –5.41 –7.066 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –5.115 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.787 –2.714 –2.714 –3.657 –3.331 –3.331 –1.107 –1.522 –1.296 –.193 –.17 0 –2.241 –3.963 –1.955 –.326 3.498 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 25.813 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.667 –8.955 –5.41 –7.268 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –4.338 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.787 –2.714 –2.714 –3.657 –3.331 –3.331 –1.107 –1.522 –.577 –.193 –.17 0 –2.241 –3.963 –1.955 –.326 3.498 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 25.813 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –6.106 –6.81 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.667 –8.955 –5.41 –7.268 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –3.46 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.115 –2.714 –2.714 –3.657 –3.331 –3.331 –1.107 –1.522 –.577 –.193 –.17 0 –2.241 –3.963 –1.738 1.939 4.328 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –6.106 –6.12 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.667 –8.955 –5.41 –7.268 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –2.926 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.115 –2.714 –3.657 –3.657 –3.331 –3.331 –1.107 –1.522 –.577 –.193 –.17 0 –2.241 –3.963 –2.578 1.939 4.328 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –6.81 –6.12 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.667 –7.748 –5.41 –7.268 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –2.926 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.115 –2.714 –3.657 –3.657 –3.331 –3.331 –1.107 –1.522 –.577 –.193 –.17 0 –2.241 –3.963 –2.578 1.939 4.85 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –6.81 –6.12 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –7.101 –7.748 –5.41 –7.268 –7.268 –7.268 –5.827 –5.115 –5.115 –5.115 –2.926 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.872 –2.115 –2.714 –3.657 –3.657 –3.331 –3.331 –1.107 –1.522 –.577 –.193 –.17 0 –2.241 –.742 –2.578 1.939 4.85 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –6.81 –6.12 –6.12 –6.12 –6.12 –6.832 –6.107 –6.107 –6.107 –6.107 –6.107 –7.101 –6.932 –5.41 –7.268 –7.268 –6.408 –5.154 –5.115 –5.115 –5.115 –2.926 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.991 –2.115 –2.714 –3.657 –3.657 –3.331 –3.331 –1.522 –1.522 –.577 –.193 –.17 0 –2.241 .445 –2.578 1.939 4.85 6.882 7.655 8.912 10.864 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –7.101 –5.683 –5.41 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –5.115 –2.926 –2.926 –2.926 –2.828 –2.828 –1.872 –1.872 –1.872 –1.991 –2.115 –2.714 –3.657 –3.657 –3.331 –3.331 –1.522 –1.522 –.577 –.193 –.17 0 –2.241 .445 –1.527 2.871 6.289 6.067 6.36 7.217 8.584 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –6.12 –6.499 –6.107 –6.107 –6.107 –6.107 –6.107 –8.955 –5.41 –5.41 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –5.115 –2.926 –2.926 –2.762 –2.828 –2.828 –1.872 –1.872 –1.872 –2.787 –2.115 –2.714 –3.657 –3.331 –3.331 –3.331 –1.522 –1.522 –.577 –.193 –.17 .249 –3.963 .445 –1.527 3.089 6.681 6.067 6.36 7.217 8.584 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.107 –8.955 –5.41 –5.41 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –5.115 –2.926 –2.926 –2.762 –2.828 –2.322 –1.872 –1.872 –1.872 –2.787 –2.115 –2.714 –3.657 –3.331 –3.331 –1.107 –1.522 –1.522 –.274 –.193 –.17 –3.184 –3.963 –1.955 –1.527 3.089 6.681 6.067 6.36 7.217 8.584 10.835 13.305 15.521 25.624 31.424 32.05 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –3.987 –5.634 –6.81 –6.12 –6.12 –6.12 –7.198 –6.107 –6.107 –6.107 –6.107 –6.107 –6.107 –8.955 –5.41 –5.41 –7.268 –7.268 –6.408 –5.115 –5.115 –5.115 –5.115 –2.926 –2.926 –3.189 –2.828 –1.872 –1.872 –1.872 –1.872 –2.787 –2.115 –2.714 –3.657 –3.331 –3.331 –1.107 –1.522 –1.522 –.274 –.193 –.17 –3.184 –3.963 –1.955 –.326 3.089 6.681 6.067 6.36 7.217 8.584 10.835 13.305 20.922 25.813 32.05 32.193 FEDERAL RESERVE BANK OF DALLAS 15 ECONOMIC REVIEW THIRD QUARTER 1995 A New Quarterly Output Measure For Texas Texas’ transition from boom to bust during the 1970s and 1980s illustrates how the Texas economy often performs differently from the nation’s. The uniqueness of the state’s economy makes it important to gather timely state-specific data to measure regional economic performance. Two frequently used measures of regional economic activity are state nonfarm payroll employment and the unemployment rate. While these measures are timely and useful, labor is only one input into the production process. Productivity, through its effect on wages and earnings, directly impacts workers’ standard of living. Output embodies the utilization and productivity of labor and capital. Therefore, it is a more comprehensive measure of economic well-being than employment measures. Analysis of output and employment data can sometimes lead to different conclusions about economic performance. For example, after peaking in 1981, Texas manufacturing employment generally declined during the rest of the decade. Manufacturing output, however, increased throughout the period. As Figure 1 shows, employment data alone could lead one to conclude that manufacturing activity was on a long-term decline, yet the output data show that this was not the case. The measurement of regional output generally has been restricted to the industrial sector, which has attracted special attention because of its strong cyclical nature and its availability of information relative to the nonindustrial sector. While timely monthly manufacturing indexes are available for several states, manufacturing represents only about 19 percent of total output, Franklin D. Berger Manager of Research Support Federal Reserve Bank of Dallas Keith R. Phillips Economist Federal Reserve Bank of Dallas A nalysis of output and employment data can sometimes lead to different conclusions about economic performance. For example, after peaking in 1981, Texas manufacturing employment generally declined during the rest of the decade. Manufacturing output, however, Figure 1 Texas Manufacturing Output and Employment increased throughout the period. Index, January 1970 = 100 250 230 210 Output 190 170 150 130 Employment 110 90 70 ’70 ’72 ’74 ’76 ’78 ’80 ’82 ’84 ’86 ’88 ’90 ’92 ’94 SOURCES OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce; Bureau of Labor Statistics, U.S. Department of Labor; Federal Reserve Bank of Dallas. 16 polate the census value-added data.2 Because census value-added data are not available for the service-producing sectors, BEA uses another method of estimation for these sectors. An alternative way to measure valueadded is to calculate the sum of payments made to the factors of production. In other words, the value added by a firm or industry can be measured by the value of labor and capital combined with intermediate inputs to produce output.3 In estimating RGDP and RGSP for the serviceproducing industries, BEA measures payments to labor and capital. Specifically, gross state product (GSP) in service-producing industries is calculated by adding: (1) employee compensation and proprietors’ income and (2) indirect business tax and nontax liability and capital charges.4 The industry totals are then deflated by the national industry implicit price deflators. on average. Fortunately, a more comprehensive measure of regional output has become available in recent years. The Bureau of Economic Analysis (BEA) of the U.S. Department of Commerce estimates nominal gross state product (NGSP) and real gross state product (RGSP). Although these data are available for sixty-one industry classifications for all fifty states and the District of Columbia, they are rarely used for current analysis or mentioned in the media because they lack timeliness and are annual. As of April 1995, the latest RGSP data available were for 1991. In this article, we estimate quarterly measures of Texas RGSP that lag the reporting quarter by about four months. For the period in which BEA’s RGSP data are available, our quarterly estimates sum to BEA’s annual figures. For the period after the BEA data, our results represent preliminary RGSP estimates that will be revised later to sum to the BEA data. Statistical measures of fit show that simple models based on changes in personal income and price indexes do well in estimating changes in RGSP at the Standard Industrial Classification (SIC) division level.1 Based on our results, Texas’ real output has grown strongly during the 1990s, although in 1993 and 1994 it grew somewhat more slowly than the nation’s. Also, Texas RGSP growth has been stronger than employment growth in the 1990s, indicating overall productivity growth of about 2 percent. A practical approach to expanding the RGSP data The long reporting lag and the data’s annual frequency severely limit the usefulness of RGSP as a timely measure of regional trends or business cycles. To increase the periodicity and timeliness of the RGSP data, we first look for timely monthly or quarterly series that might move in a fashion similar to BEA Releases 1992 GSP Data RGSP. Using standard statistical techniques, BEA released 1992 GSP data shortly before we examine the relapress time for this article. While unable to incorporate the new data fully into our analysis, we are able to tionship between the check the accuracy of our 1992 forecasts. On the annualized candidate whole, the magnitude of the errors is consistent with series and RGSP and the errors estimated for 1990 and 1991. The out-ofuse these results to insample forecasting results for RGSP for 1992: terpolate RGSP at a Industry Percent error higher frequency and to extrapolate RGSP Goods-producing sectors forward in time. Agriculture 3.8 Mining – 5.5 Knowledge of Construction 9.2 RGSP’s construction Durable manufacturing – 2.3 provides insight into Nondurable manufacturing – 4.0 possible data series Service-producing sectors and techniques to conTransportation, communication, struct timely monthly and public utilities 1.7 or quarterly RGSP Wholesale trade –.7 measures.5 As previRetail trade –.9 Finance, insurance, ously described, inand real estate 1.5 dustry-level RGSP is Services –.6 constructed differently Government .9 for service-producing Total RGSP .2 industries than for goods-producing (manWeighted sum of absolute errors 2.1 ufacturing, mining, and construction) in- What is RGSP? RGSP is the regional equivalent of real gross domestic product (RGDP) as reported in the national income and product accounts. To avoid double-counting, industry-specific RGSP is measured so that the sum of RGSP across all industries equals total real output. That is, each industry’s RGSP is a measure of value-added and is different from the total number of units produced or the total sales of an industry. One way to measure value-added is to calculate the gross market value of the goods and services produced by an industry and subtract the value of intermediate products and services purchased. BEA uses this method to calculate NGSP for the goods-producing sectors. To estimate NGSP in the goods-producing industries, BEA subtracts an estimate of purchased services from the estimates of value-added reported by the Census Bureau. To construct RGSP, BEA deflates these series by national industry-specific implicit price deflators. For noncensus years, BEA uses the Annual Survey of Manufactures (ASM) and other data to interpolate and extra- FEDERAL RESERVE BANK OF DALLAS 17 ECONOMIC REVIEW THIRD QUARTER 1995 Table 1 Composition of Texas Gross State Product, 1991 Industry Goods-producing sectors Agriculture Mining Construction Durable manufacturing Nondurable manufacturing Service-producing sectors Transportation, communication, and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services Government Labor costs/GSP Industry RGSP/ Total RGSP .90 .46 .94 .76 .48 .018 .073 .037 .082 .078 .53 .64 .66 .120 .070 .098 .36 .90 .96 .153 .162 .107 the data are interpolated and extrapolated using wages and salaries from the personal income data.7 Although information on capital utilization generally is not available or is costly to obtain on a timely basis, the lack of it may not be a significant impediment to estimating RGSP in the service-producing sectors. One reason is that personal income is the basis for much of the year-to-year movement in capital charges. Another reason is that the service-producing sectors are generally labor intensive. As Table 1 shows, the share of value-added represented by the labor component is above 60 percent in the service-producing sectors, with the exceptions of the finance, insurance, and real estate (FIRE), and transportation, communication, and public utility (TCPU) industries. In services and government —which together represent slightly more than 25 percent of RGSP—labor’s share is 90 percent or more. For changes in the labor component of RGSP to be a good representation of changes in total RGSP, the variance of the labor component should be high relative to the variance of the capital component, or the movements in the labor and capital components should be highly correlated, or both.8 The variance decomposition of RGSP in Table 2 shows that, for most industries, the variance of RGSP is due mainly to the variance of the labor component and the covariance between labor and capital. This is particularly true for the government and for service sectors in which the capital component has varied little over time. The main exception is the FIRE sector. Overall, the variance decomposition of RGSP suggests that, for most service-producing industries in Texas, extrapolating RGSP solely on the basis of changes in the labor component is worthwhile. As mentioned earlier, BEA estimates RGSP in the goods-producing industries using a different approach. For farming, mining, construction, and manufacturing, BEA estimates RGSP directly, using census data on value-added in production. For farming, mining, and construction in the noncensus years, BEA estimates RGSP mainly using changes in earnings from the personal income data. This method suggests that, for most years, changes in labor income should be a good representation of changes in RGSP in these industries. The results in Tables 1 and 2 also suggest that changes in the labor component could be useful in approximating changes in total RGSP for the agriculture and construction industries. The variance of capital is relatively high for mining, and the absolute value of the covariance SOURCE OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce. dustries. The difference in construction and the general availability of more monthly and quarterly series relating to the goods-producing sectors warranted a separate investigation into the estimation of RGSP in these two sectors. We start with a discussion of the service-producing industries. RGSP in the service-producing industries is calculated by summing the factor payments to labor and capital and dividing this total by the national implicit price deflator for the industry. BEA’s estimates of payments to labor (primarily employees’ compensation and proprietors’ income) come mostly from state personal income data also produced by BEA. For example, in 1987 personal income data represented 93 percent of the employees’ compensation and proprietors’ income components of GSP.6 Because state personal income data are available quarterly at the SIC division level and represent most of the labor component of GSP, these data are a likely candidate for estimating nonindustrial output on a more timely basis. RGSP’s nonlabor component comprises primarily sales and property taxes levied by state and local governments, corporate profits with inventory valuation adjustment, corporate capital consumption allowances, business transfer payments, net interest, rental income of individuals, and subsidies less the current surplus of government enterprises. For the census years 1977, 1982, and 1987, much of the information for estimating nonlabor charges for the service-producing industries comes from various censuses and company-specific data reported by various regulatory agencies. For noncensus years, much of 18 Table 2 Variance Decomposition of Texas Real Gross Product Variance of RGSP = variance of labor component + variance of capital component + 2 ⫻ covariance Industry Total variance Labor variance Capital variance 2 ⫻ covariance Goods-producing sectors Agriculture Mining Construction Durable manufacturing Nondurable manufacturing 789.5 5,548.3 7,474.5 8,320.4 16,332.5 1,370.2 2,945.5 4,936.5 5,796.3 871.4 271.8 4,550.1 618.1 921.1 13,040.5 – 852.5 –1,947.3 1,919.9 1,603.0 2,420.6 23,715.1 15,334.5 21,268.0 6,596.3 5,796.5 6,320.4 6,508.5 2,345.5 4,857.7 10,610.3 7,192.5 10089.9 39,903.2 60,862.7 9,667.5 3,389.1 48,562.8 8,296.9 22,529.7 716.9 106.4 13,984.5 11,583.1 1,264.2 Service-producing sectors Transportation, communication, and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services Government SOURCE OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce. several industry-specific and general price deflators to determine which—when combined with the personal income data—have the greatest ability to explain changes in RGSP. suggests that only a small portion of the changes in the capital component can be accurately predicted by changes in the labor component. BEA uses state-level value-added data from the ASM to estimate manufacturing RGSP in the nonbenchmark years. Thus, from a pragmatic approach, it is unclear if the personal income data would be a good representation of RGSP in the manufacturing sector. Labor’s low factor share and its relatively low contribution to the variance of nondurable manufacturing RGSP, as Tables 1 and 2 show, also indicate that the labor component may be a poor predictor of nondurable manufacturing RGSP. Fortunately, for durable and nondurable manufacturing, electric power usage data are available to proxy capital usage.9 Finally, determining how to account for price changes is an important issue in using personal income data to estimate RGSP. As explained earlier, BEA deflates nominal GSP by national industry-specific implicit price deflators to calculate RGSP. Implicit price deflators are simply the ratio of nominal to real gross product originating. Real gross product originating is derived by separately deflating the value of production and the cost of materials. It is not apparent whether changes in the implicit price deflators would be more closely tied to changes in industry-specific price deflators or to changes in more general price deflators such as the consumer price index (CPI). Therefore, we examine FEDERAL RESERVE BANK OF DALLAS The model The procedure we use to distribute RGSP across quarters within-sample and to extrapolate RGSP out-of-sample is the method of best linear unbiased interpolation and extrapolation, introduced by Chow and Lin (1971).10 A key feature of the Chow–Lin procedure is the restriction that the quarterly in-sample values sum to the annual data. Prior to running the procedure, we run OLS regressions to test the appropriate dynamics of the equations. OLS regressions of the following form have been run for each SIC division: ln(RGSPit ) = β0 + β1ln(Eit ) – β2ln(Pit ) + et , where E is earnings (wages and salaries, other labor income such as employer contributions to privately administered pension and welfare funds, employer contributions for social insurance, and proprietors’ income with inventory valuation) from the personal income data; P is the price deflator used for the industry; i and t are industry and time subscripts; the betas are estimated coefficients; and ln refers to the natural log of the series. We run the equation on annual data from 1969 to 1989 and test the errors, et , 19 ECONOMIC REVIEW THIRD QUARTER 1995 Table 3 Summary Measures of Volatility and Model Fit to the annual RGSP data. This procedure allows the model’s dynamics to Personal income Employment Variance of be correctly specified while restricting Industry model model growth rates the quarterly in-sample series levels to Goods-producing sectors sum to the annual data. Agriculture .716 N.A. .025 We perform this procedure on Mining .192 –.054* .008 each of the eleven SIC divisions. For Construction .750 .645 .007 Durable manufacturing .805 .751 .007 the durable and nondurable manufacNondurable manufacturing .378 .023* .005 turing equations, electric power usage data are included as a measure of capiService-producing sectors Transportation, communication, tal usage. The Durbin–Watson statisand public utilities .473 .479 .001 tics from the differenced regressions Wholesale trade .394 .295 .003 show little evidence of autocorrelation Retail trade .749 .384 .002 so no adjustment to the errors was Finance, insurance, performed. The F-statistics from the and real estate .377 .194 .004 Services .829 .248 .0004 regressions are all significant, and the Government .299 –.023* .0002 adjusted R 2 s show strong predictive power. Although the information in N.A. = not applicable. Tables 1 and 2 suggests that the per* The equation is not statistically significant at the 5-percent level. sonal income data would be a good SOURCES OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce; Bureau of Labor predictor of changes in RGSP, we exStatistics, U.S. Department of Labor. amine another model that avoids the necessity of using price deflators in estimating RGSP. for stationarity with the Augmented Dickey– Payroll employment is available for the Fuller (ADF) test.11 In the levels form of the nonagricultural industries we have studied; equation, we find the errors to be nonstationary therefore, an alternative method of estimating across all industries, suggesting that the models RGSP is to estimate labor productivity by industry be run in first differences. Because the small and multiply these estimates by the employnumber of observations reduces the reliability of ment data.12 The model we estimate is the ADF tests, we make out-of-sample com⎛ RGSPit ⎞ parisons using the Chow–Lin procedure on both ∆ ln⎜ ⎟ = β0 + β1∆ ln EMPit + eit , the levels and differenced forms of the equation. ⎝ EMPit ⎠ The mean weighted out-of-sample errors for 1990 and 1991 are smaller for the differenced where ∆ indicates first differences and EMP is equations than for the levels equations —further nonagricultural employment. β0 represents the evidence that the differenced form of the model long-run productivity growth rate, and β1 repis appropriate. resents the relationship between employment Series differences have been calculated as and productivity. This equation is run for each the natural log of the series minus the natural log SIC division, except agriculture, using the Chow– of the series four quarters earlier. The Chow–Lin Lin procedure. By first adding the natural log of procedure performed on the differenced data employment to both sides of the equation, this creates a quarterly estimate of the percentage model’s fit can be compared with the fit of the change in RGSP by industry. To transform these personal income model. As previously stated, changes into levels, the Chow–Lin procedure for the durable and nondurable manufacturing initially is performed on the levels of the equations, electric power usage is included as a data, and the quarterly level estimates for 1969 measure of capital usage. are used with the series of estimated changes As Table 3 shows, the adjusted R 2 s from to estimate industry output during the entire the employment model are generally much period. These RGSP estimates do not exactly lower than the adjusted R 2 s from the personalsum to the actual annual RGSP estimates. To income model. The main exception is the ensure that the quarterly estimates sum to TCPU industry, which has a slightly better fit the annual RGSP data, we treat these estimates using the employment model. as independent variables and use them in the Chow–Lin procedure with the annual RGSP data. Results This treatment ensures that the final quarterly To evaluate the out-of-sample performin-sample RGSP estimates are restricted to sum ance of our estimates, only data through 1989 Adjusted R 2 20 errors and the errors for total RGSP are generally low for the two years. The mean weighted absolute error is 2.2 percent for 1990 and 3.9 percent for 1991. The error for total RGSP is 0.6 percent for 1990 and –1.6 percent for 1991. When one considers that the nation was in recession in parts of 1990 and 1991, the model seems to perform well, at least in the aggregate. To calculate our final RGSP estimates, we rerun the Chow–Lin procedure and include the data through 1991 and calculate out-of-sample estimates for the period from first-quarter 1992 through fourth-quarter 1994. Figure 2 shows that while Texas employment declined only slightly during the national recession from July 1990 to March 1991, Texas RGSP declined for two consecutive quarters. Thus, the RGSP data suggest that the Texas economy was weaker during this period than the employment data indicate. Figure 2 also shows that during the 1990s real output growth has outpaced employment growth, indicating an overall increase in labor productivity of about 2 percent.13 During the 1990s, real output growth has been stronger in Texas than in the nation, al- Figure 2 Texas Employment and Real Gross Product Index, 1990:1 = 100 116 114 112 RGSP 110 108 Employment 106 104 102 100 98 1990 1991 1992 1993 1994 SOURCES OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce; Bureau of Labor Statistics, U.S. Department of Labor. are included in the regressions. The out-ofsample errors give additional information on the model’s performance by simulating how the model would have performed had we used it prior to the availability of the 1990 and 1991 RGSP data. As Table 3 shows, the out-of-sample errors vary across industries, with goodsproducing industries generally experiencing the largest errors. These out-of-sample errors are consistent with the in-sample measures of fit (Table 4 ). Although the adjusted R 2 s for the agriculture and construction industries show that the model explains a fairly large percentage of the fluctuations in growth in these industries, variance measures show that these industries are particularly volatile. Because of the large out-of-sample errors in the agriculture, mining, and construction industries, we experiment with adding real production measures to the regressions for these industries. For example, when we add the number of residential permits and the square feet of nonresidential permits to the construction equation, we find the coefficients of these measures to be jointly statistically insignificant. Similarly, the addition of a measure of agricultural production to the equation for this industry, and oil and gas production was added to the mining equation yields no significant increases in fit for either industry. The main source for the large error for nondurable manufacturing in 1991 is a very large drop in reported RGSP for the chemicals industry. This large drop is inconsistent with personal income and employment data for that industry. Although several industries experience large out-of-sample errors, the average absolute FEDERAL RESERVE BANK OF DALLAS Table 4 Out-of-Sample Forecasting Results for RGSP, 1990 and 1991* Percent error Industry Goods-producing sectors Agriculture Mining Construction Durable manufacturing Nondurable manufacturing Service-producing sectors Transportation, communication, and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services Government Total RGSP Weighted sum of absolute errors 1990 1991 Deflator used 11.7 12.0 4.1 .7 1.6 14.0 1.9 10.2 1.8 –20.8 Agriculture PPI Mining PPIs Total CPI Total CPI Total CPI –1.2 –4.5 –1.0 3.0 –4.0 –2.4 TCPU CPI Total CPI Total CPI –1.5 –.03 –.3 –3.1 –.1 –.1 Total CPI Services CPI Total CPI .6 –1.6 2.2 3.9 * The model used is equation 2 in the text, in which the variables are in first differences of natural logs. We use the models’ estimates of quarterly changes to estimate quarterly log levels by the method described in the text. The quarterly log levels are exponentiated and summed to produce an estimate of annual RGSP. The percentage difference between the annualized estimate and actual RGSP is shown in the table. A negative number indicates an overestimate of RGSP, while a positive number indicates an underestimate. 21 ECONOMIC REVIEW THIRD QUARTER 1995 though over the past two years this has not been the case, as shown in Figure 3. The relative strength of national output growth in recent years has come from large gains in labor productivity; employment growth was faster in Texas during both years. Manufacturing and construction output in Texas accelerated in 1994 after weakness in the early 1990s (Figure 4 ). Output growth in most of the service-producing industries has been strong throughout much of the 1990s (Figure 5 ). The trade and TCPU industries have performed the strongest, while the government and FIRE industries have been weak. Figure 3 Texas and U.S. Real Gross Product Index, 1990:1 = 100 116 114 Texas 112 110 108 106 United States 104 102 100 98 1990 1991 1992 1993 Summary and conclusion 1994 Giese (1989) states that “the important contribution of BEA’s GSP data is that they provide a more accurate and comprehensive measure of regional output than other regional data.” Although RGSP can be very useful to the regional analyst, its main drawbacks are its annual periodicity and lack of timeliness. In this article, we set out to improve the RGSP data for Texas by increasing its periodicity and timeliness. The method we use is best linear unbiased distribution and extrapolation, developed by Chow and Lin (1971). We find that the Chow–Lin procedure in first-difference form using personal income and various price measures does quite well in out-ofsample forecasts for 1990 and 1991. We use the procedure to produce RGSP data for each SIC division through the fourth quarter of 1994 and show that real output in the state has not grown as fast as in the United States over the past two years. The data developed in this article are available by accessing Dallas Fed’s free electronic bulletin board—Fed Flash— at (214) 922-5199 or (800) 333-1953. The new quarterly output measures should enhance analysts’ ability to understand current economic conditions in Texas. SOURCE OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce. Figure 4 Texas Construction and Manufacturing Output Index, 1990:1 = 100 116 112 Manufacturing 108 104 100 Construction 96 92 1990 1991 1992 1993 1994 SOURCE OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce. Figure 5 Texas Service-Producing Industries’ Output Index, 1990:1 = 100 132 128 Wholesale trade Notes 124 120 116 Retail trade 112 TCPU* 108 1 Services Government 104 100 FIRE** 96 2 92 1990 1991 1992 1993 1994 * Transportation, communication, and public utilities. ** Finance, insurance, and real estate. 3 SOURCE OF PRIMARY DATA: Bureau of Economic Analysis, U.S. Department of Commerce. 22 The authors thank Steve Brown, Bill Gilmer, Lori Taylor, and D’Ann Petersen for helpful comments. The SIC-division-level industries are agriculture; construction; mining; durable goods manufacturing; nondurable goods manufacturing; finance, insurance, and real estate; services; retail trade; wholesale trade; transportation, communication, and public utilities; and government. For more information about the calculation of GSP, see Beemiller and Dunbar (1993); Trott, Dunbar, and Friedenberg (1991); and Giese (1989). Strictly speaking, the exhaustion of nominal valueadded by payments to the factors of production 4 5 6 7 8 9 10 11 12 13 requires the assumption of linear homogeneous production functions and perfectly competitive labor markets. While recognizing that the usage may not be precise, for the purposes of this article all nonlabor payments will be referred to as capital payments. Although BEA also calculates these categories for the goods-producing sectors, total gross product for the goods-producing sectors is not calculated as the sum of these four categories but is based on census valueadded data. In the goods-producing industries, the capital component is estimated as the residual of total gross product minus the other components, which are measured directly. References Before BEA began producing the GSP data in 1988, many regional analysts used the Kendrick–Jaycox (K–J) methodology to estimate GSP. Essentially, K–J methodology allocates GDP (by industry) to the states by using each state’s earnings’ share of total U.S. earnings. The availability of the BEA data essentially makes the K–J method obsolete. For a comparison of the BEA data to estimates calculated with the K–J methodology, see Giese (1989). Most of the difference is employers’ contributions to social insurance, which come from another source. For more information on the sources of the capital estimates, see the table on page 36 of Beemiller and Dunbar (1993). The higher the absolute value of the covariance between the labor and capital components (for given variances in the labor and capital components), the less information is lost by estimating RGSP with just the labor component. For example, if labor and capital were perfectly correlated, then one could calculate RGSP using some constant multiple of the labor component. Previous research validates the use of electric power consumption as a proxy for capital usage (Moody 1974). The authors wish to thank Jeffery W. Gunther for transforming Chow and Lin’s exposition into working computer code. For more information on testing for stationarity in the residuals using the Augmented Dickey–Fuller (ADF) test, see Engle and Yoo (1987). A variant of this method would be to use estimates of U.S. productivity by industry to proxy Texas productivity. Although it would be interesting to test the ability of this method, U.S. productivity estimates are not available with the necessary industry detail, timeliness, and periodicity. In calculating productivity growth, it was assumed average weekly hours worked remained constant over this period. Also, the employment data do not include the agricultural sector. A comparison of growth in nonfarm RGSP with growth in the nonfarm employment data gives approximately the same productivity growth as indicated in Figure 1. Chow, Gregory C., and An-loh Lin (1971), “Best Linear Unbiased Interpolation, Distribution, and Extrapolation of Time Series by Related Series,” Review of Economics and Statistics 53 (4): 372 – 75. FEDERAL RESERVE BANK OF DALLAS Beemiller, Richard M., and Ann E. Dunbar (1993), “Gross State Product, 1977– 90,” Survey of Current Business 73 (December): 28 – 49. Berger, Franklin D., and William T. Long III (1989), “The Texas Industrial Production Index,” Federal Reserve Bank of Dallas Economic Review, November, 21– 38. Board of Governors of the Federal Reserve System (1986), Industrial Production, 1986 Edition, Washington, D.C., vii. Engle, Robert F., and Byung Sam Yoo (1987), “Forecasting and Testing in Co-Integrated Systems,” Journal of Econometrics 35 (May): 143 – 59. Giese, Alenka S. (1989), “A Window of Opportunity Opens for Regional Economic Analysts: BEA Releases Gross State Product Data,” Federal Reserve Bank of Chicago Working Paper Series Regional Economic Issues, 3. Israilevich, Philip R., Robert H. Schnorbus, and Peter R. Schneider (1989), “Reconsidering the Regional Manufacturing Indexes,” Federal Reserve Bank of Chicago Economic Perspectives, July/August, 13 – 21. Kenessey, Zoltan (1994), “Regional Monthly Production Indexes in the United States,” in Forecasting Financial and Economic Cycles, by Michael P. Niemira and Philip A. Klein (New York: John Wiley and Sons, Inc.) 329 – 46. Moody, Carlisle E. (1974), “The Measurement of Capital Services by Electrical Energy,” Oxford Bulletin of Economics and Statistics 36 (February): 45 – 52. Trott, Edward A., Ann E. Dunbar, and Howard L. Friedenberg (1991), “Gross State Product by Industry, 1977– 89,” Survey of Current Business 71 (December): 43 – 59. 23 ECONOMIC REVIEW THIRD QUARTER 1995 Alternative Methods Of Corporate Control in Commercial Banks This article investigates the corporate control mechanism that operates in commercial banks. The term corporate control mechanism refers to the various methods by which bank owners attempt to force bank management to follow value-maximizing policies. Various devices can motivate such managerial discipline. External devices—the market for takeovers, external capital, and the final output of the firm— can all in theory discipline managers by threatening them with replacement or bankruptcy of their firm. Internal devices consist of direct monitoring performed by boards of directors and large shareholders and the management compensation contract, which can provide incentives to maximize value by giving managers equity-like shares in the firm. This article analyzes the use of some of these corporate control devices in banks. Although the research on the corporate control mechanism in nonfinancial firms is vast, there is surprisingly little research on the corporate control mechanism operating in banks. Yet analysis of the corporate control mechanism in banks is important for a number of reasons. First, despite its supposed decline in recent years, banking remains an extremely important industry that acts as the main interface between savers and investors. Second, such analysis contributes to our understanding of the different ways in which corporate control mechanisms operate in firms under different legal and regulatory environments. The considerable differences between the legal and regulatory environment of banks and nonfinancial firms may imply substantial differences in the nature and effectiveness of their respective corporate control mechanisms. In particular, federal and state restrictions on the market for corporate control for banks and the oligopolistic advantages that commercial banks have in issuing insured debt may mean that important external market mechanisms for disciplining managers—the takeover and product market—are significantly weaker for banks. The regulatory environment of the commercial banking industry may substitute to some degree for the weaker market mechanisms of corporate control. However, intervention by the regulatory authorities is widely regarded as a poor, more costly substitute for market control mechanisms, both because of bureaucratic and political problems that interfere with the efficient functioning of regulatory agencies and because maximizing shareholder value (the objective of market mechanisms) is not the same as minimizing the probability of failure (the regulator’s objective). This article addresses the question of whether Stephen D. Prowse Senior Economist and Policy Advisor Federal Reserve Bank of Dallas A lthough the research on the corporate control mechanism in nonfinancial firms is vast, there is surprisingly little research on the corporate control mechanism operating in banks. Yet analysis of the corporate control mechanism in banks is important for a number of reasons. 24 these differences in the regulatory environment of banks relative to nonfinancial firms have produced greater reliance on internal devices for corporate control—active boards and large, active shareholders—or, if not, whether the corporate control problem is simply more severe in commercial banking. Third, such analysis may provide information on whether commercial banks suffered from a corporate control problem in the 1980s, as some researchers have recently proposed (Gorton and Rosen 1992). Many analysts claim that over the past ten to fifteen years, the U.S. commercial banking industry has suffered a significant decline in performance, including a loss in market share to nonbank competitors (such as securities markets, mutual funds, insurance companies, finance companies, and foreign banks), substantial falls in bank profitability, and a skyrocketing bank failure rate.1 All this has occurred despite intense merger and acquisition activity among banks that was supposed to improve productivity and cost efficiency. Many researchers believe that the reasons for this decline are secular in nature and that the recent recovery in bank profitability will prove to be only a temporary phenomenon, with commercial banking continuing to decline relative to other financial institutions over the long term. Researchers have proposed numerous reasons for the commercial banking industry’s woes in the 1980s. Greater competition from nonbanks and a heavier federal regulatory burden are often put forward as reasons for this apparent decline.2 Others point to the moral hazard problems that appear particularly severe in the banking industry.3 This article addresses another possible reason for the relative underperformance of banks: that the corporate control mechanism in commercial banks is less effective than in nonbank firms. Finally, from a public policy viewpoint, examination of the corporate control mechanism in banks may be useful in evaluating the industry’s current legal and regulatory environment and also some of the recently proposed banking legislation that may amend or eliminate provisions in the Glass–Steagall Act. While much of the current and proposed legislation has been evaluated in terms of the desirability of allowing commercial banks to engage in securities underwriting or in selling insurance, there has been little analysis in terms of the effects on the corporate control mechanism that operates in banks, even though some of the proposed changes in banking law would loosen the restrictions on bank ownership, with potential effects FEDERAL RESERVE BANK OF DALLAS both on the structure of bank ownership and the bank takeover market. In this article, I attempt to provide such analysis. I analyze the corporate control mechanism in U.S. commercial bank holding companies (BHCs) over the period 1987–92 using data on the number of managers versus outsiders on a BHC’s board of directors; the ownership structure of the BHC, including directors’ shareholdings and the stakes of the BHC’s largest shareholders; and various measures of bank performance. I relate these variables to five types of corporate control change a BHC could undergo over the sample period: hostile takeover, friendly acquisition, removal of top management by the board of directors, intervention by regulators, and no control change. I use these data to examine the relative importance and effectiveness of the different methods of disciplining managers in BHCs and how they differ from those employed in nonfinancial firms. Some questions this article addresses are, What are the primary means by which managers are disciplined in commercial banks? What is the frequency and effectiveness with which these means are used? For example, what is the frequency of top management turnover in commercial banks? Is turnover related to measures of bank performance? How important are boards of directors in disciplining top management relative to alternative control devices such as hostile takeovers, friendly acquisitions, and intervention by regulators? What is the structure of ownership in commercial banks, and is it related to bank performance? As mentioned above, many of these questions have been addressed for U.S. nonfinancial firms (see, for example, Morck, Shleifer, and Vishny 1989 and Jensen and Murphy 1990), so some standards are available with which results for the banking sector can be compared. This study borrows in particular the method employed in Morck, Shleifer, and Vishny (1989) for their sample of manufacturing firms. In the next section of this article, I outline the factors that are unique to the commercial banking sector that may affect the nature and the effectiveness of its corporate governance mechanism, and I survey the academic research on corporate governance problems in commercial banks. The subsequent section describes the data and discusses the empirical results. The final section concludes. The corporate control mechanism in commercial banks Does the legal and regulatory environment of U.S. commercial banks today imply a 25 ECONOMIC REVIEW THIRD QUARTER 1995 system of corporate governance different from that observed in other sectors of the economy? Many unique factors in the commercial bank operating environment may influence the nature and effectiveness of the corporate control mechanism in commercial banks. The first unique factor is federal regulation of the takeover market. The threat of a takeover of a firm, in which management usually is replaced, can discipline managers to act in the interests of shareholders. Restrictions on the type or number of potential acquirers of the firm make takeovers less likely and thus limit the credibility of the takeover threat. In the banking sector, there traditionally have been significant restrictions on the takeover market. For example, the Bank Holding Company Act (as amended in 1970) and the National Banking Act generally require that the acquirer of a commercial bank also be a commercial bank or bank holding company—mergers between nonbank corporations and commercial banks are prohibited—and there are more general restrictions on the ownership of banks by nonfinancial corporations. In addition, federal regulation may make permitted hostile takeovers within the commercial banking sector much more expensive and time consuming than in nonbank sectors of the economy. Interstate banking regulations may, for example, prohibit many possible bank mergers. In addition, bank takeovers typically face extensive delays. This tendency may lower the frequency of hostile takeovers, which typically depend for their success on the ability to close the transaction quickly. Bank takeovers require prior approval from one of the three federal bank regulators —the Comptroller of the Currency, the Federal Deposit Insurance Corporation (FDIC), or the Federal Reserve Board—and state authorities (Baradwaj, Fraser, and Furtado 1990). After approval is granted, there is a thirtyday waiting period so the Justice Department can scrutinize the takeover attempt. In all, the takeover process can last four months or longer. In many cases, these restrictions may make the threat of a takeover in commercial banking insufficient to discipline managers. Such restrictions may also influence the ownership structure of commercial banks. Currently, nonfinancial corporations and firms in important financial sectors such as the insurance industry are prohibited from owning commercial banks. To a large extent, the law restricts ownership of commercial banks to individuals and other commercial banks. To the degree that this restriction reduces the likelihood that banks will have equity holders with large stakes at risk, it also may reduce the effectiveness of one mechanism of corporate control: the monitoring and oversight performed by shareholders motivated by their large holdings. Another unique factor is the effect of deposit insurance on the moral hazard problem in banking. As is the case with any limited liability firm with debt outstanding, bank stockholders have incentives to take on inefficient risk. However, the problem is more acute in commercial banks, where stockholders are in addition subject to the distorting incentives arising from the existence of fixed-price deposit insurance premiums. These premiums result in a subsidy to bank shareholders that increases in value with the riskiness of the bank. Thus, bank shareholders have even stronger incentives to take on inefficiently risky investments that benefit themselves at the expense of the deposit insurance fund and the taxpayers who back the fund.4 Competition in the product market can play a role in reducing the extent to which managers shirk from value maximization goals. Together with thrifts, credit unions, and government-sponsored enterprises, commercial banks have traditionally had strong oligopolistic advantages on the liabilities side of their business— the issuance of insured debt. This oligopolistic position may have given banks the scope to be more inefficient in some aspects of their business—for example, in the degree to which managers follow value-maximizing policies— yet still be competitive with other financial institutions that have not had the benefit of issuing liabilities backed by a federal guarantee. However, the advantages from issuing insured debt for banks likely have declined over recent years with the emergence of numerous good substitutes, such as money market mutual funds. Federal regulation and moral hazard clearly play a role in shaping the corporate control mechanism that operates in banks and in particular are likely to make it operate significantly differently from the corporate control mechanism at work in other firms. Nevertheless, there is only a relatively small amount of literature, particularly of recent vintage, that attempts to document empirically the existence of corporate control problems between bank shareholders and managers. Much of this work uses data from the 1970s and earlier and thus has an uncertain relevance to the banking industry as it now is configured.5 Gorton and Rosen (1992) and Allen and Cebenoyan (1991) both present evidence on the behavior of commercial banks in the 1980s that is consistent with a corporate 26 control problem. Allen and Cebenoyan find that banks with entrenched management tend to engage in the most active acquisition programs, consistent with the view that such programs are designed to increase the perquisites available to management (which vary directly with the size of the firm) rather than to increase profitability. Gorton and Rosen present evidence that entrenched managers may be a more important problem in banking than the moral hazard associated with deposit insurance. The authors find that banks that are characterized as having managements that are relatively free from outside shareholder control make the riskiest and most unprofitable investments. While both Allen and Cebenoyan and Gorton and Rosen find evidence of a corporate control problem in banks in the 1980s, neither study identifies the aspects of commercial banks’ corporate control mechanism that may be deficient or why these deficiencies may occur. This article attempts to provide an initial pass at such an analysis by examining the frequency of different types of corporate control change among BHCs in the late 1980s and their relationship with the ownership, board structure, and performance of the BHC. Data Corp.’s Mergers and Acquisitions Database. Four appear to have started as hostile takeovers and twenty-five as friendly mergers. Following MSV, I record an acquisition as hostile if the initial bid for the target was unsolicited and not accepted by the board in its initial form. Targets that were not classified as hostile were recorded as friendly. Hostile takeovers almost by definition involve changes in current management and therefore can be viewed as a change in corporate control. The degree to which friendly mergers can be so regarded is somewhat more doubtful. The fact that a friendly merger offer is not contested by current management may mean managers believe their jobs are secure. However, this belief may not prove true. In any case, the acquiring firm may keep current management but force it to make policy changes that it otherwise would not have made. For these reasons, I consider friendly mergers as potential mechanisms of corporate control change, although of a different nature than hostile takeovers. I attempt to classify those BHCs in my sample that have experienced a top management turnover. Again, following MSV, I define management turnover as a complete change between 1987–92 in the list of officers signing the letter to shareholders in the annual report. A BHC experiences a management turnover if none of the officers who signed the annual report in 1992 also signed five years earlier. I consider such turnover to be the result of disciplinary management changes forced by the board of directors.6 A BHC that has experienced a management turnover prior to being acquired is classified as an acquisition, not a turnover. This happens in four cases, in each of which the subsequent merger is friendly. As MSV note, while the board is arguably trying to deal with management problems, the BHC’s subsequent acquisition is evidence that the board’s action is not providing an adequate solution. This definition of top management turnover yields twenty-four cases of management turnover.7 The final category of corporate control change I consider is intervention by regulators. Intervention may be viewed as a “last resort” mechanism for those BHCs that may or may not have undergone previous corporate control changes yet have continued to perform poorly. Each federal banking agency, as well as each state banking authority, can impose a broad range of enforcement actions on management. Both formal and informal regulatory enforcement actions are a response to poor performance by the BHC in some aspect of its opera- Data and empirical results Frequency of corporate control changes. I analyze the frequency with which corporate control changes occur in a sample of BHCs over the period 1987–92 and the relative importance of those corporate control mechanisms that precipitate such action, such as hostile takeovers, other mergers, internally driven board turnover of the management team, and intervention by regulators. To analyze the frequency of alternative control changes, I follow the Morck, Shleifer, and Vishny (1989) (MSV) method in their study of Fortune 500 manufacturing firms. I collected data on the following characteristics of BHCs that existed in 1987: accounting data from COMPUSTAT (from 1987–92) and stock return data from the CRSP tapes (from 1983–86). In addition, I collected data on the composition of the BHC’s board of directors between insiders and outsiders and their shareholdings in 1987 and on the shareholdings of greater than 5-percent owners of the BHC in 1987 from the 10–K, annual report, or other Securities and Exchange Commission (SEC) filings. I was left with 234 BHCs in the sample, including all the largest ones. Of the 234 BHCs in the sample, twentynine were acquired by third parties during 1987– 92, based upon an examination of Securities FEDERAL RESERVE BANK OF DALLAS 27 ECONOMIC REVIEW THIRD QUARTER 1995 tions. These actions involve directing current management to attain specific capital ratios, suspend dividends, rectify loan quality problems, address liquidity and concentration problems, and the like. They can therefore be seen as a last-resort, nonmarket-based external mechanism of management discipline. Since some informal enforcement actions are never made public, there is a problem in identifying those BHCs that are subject to regulatory intervention.8 One solution would be to use the BOPEC rating—the rating assigned to the BHC by regulators —and to assume that those BHCs rated unfavorably were subject to some form of regulatory intervention.9 An alternative is to use data on the bad loans outstanding at BHCs. In this article, I construct the regulatory intervention group by ranking my sample of BHCs according to the percentage of total assets that are in the form of nonperforming or greater than ninety days past due loans.10 If a BHC was in the bottom decile of my sample in any one year of the sample period, I assume that BHC comes under regulatory intervention starting in that year.11 This definition yields thirtythree cases of regulatory intervention. BHCs that underwent a management turnover before being observed in the bottom decile of the bad loan ratio are classified as being in the regulatory intervention category, not the turnover category. Again, the argument is that while the board may be trying to deal with management problems, subsequent intervention by regulators is evidence that the board’s action is not an adequate solution. This happens in six cases. Table 1 lists the frequency of these various corporate control events, with those of the MSV study of manufacturing firms as a standard of comparison. First note that, in terms of percentages of the sample size, total corporate control changes (defined to include intervention by regulators for the BHC sample) appear to be about as frequent among BHCs as they are among manufacturing firms.12 However, the composition of total control changes between the various alternatives differs dramatically between the two groups. Market-based corporate control changes (excluding control changes owing to regulatory intervention) are about two-thirds as frequent among the sample of BHCs as they are for nonfinancial firms.13 If my measure of the regulatory intervention group does not overstate the number of BHCs subject to regulatory intervention in this period, it appears that the primary mechanism of corporate control change among BHCs in this period was in fact intervention by regulators.14 Looking at the relative frequency of the market-based control mechanisms —which is invariant to the size of the regulatory intervention group —while friendly mergers are slightly more frequent among the BHC sample, hostile takeovers and management turnover are markedly less frequent. For example, MSV record forty hostile takeovers representing 8.8 percent of their sample of nonfinancial firms. Similarly, 20.5 percent (ninety-three cases) of their sample undergo an internally precipitated management turnover. In my sample of BHCs, only 1.7 percent (four cases) undergo a hostile takeover, while 10.2 percent (twenty-four cases) of the sample undergo a management turnover.15 Thus, hostile takeovers are over five times more frequent among manufacturing firms than among BHCs, confirming the conventional wisdom. In addition, however, management turnover by the board appears twice as frequent in nonfinancial firms as in BHCs. Thus, the lower frequency of hostile takeovers among BHCs does not appear to be reflected in a greater tendency by boards to remove management at BHCs than at manufacturing firms.16 Indeed, boards at BHCs appear to be less active in removing management for disciplinary reasons. The following sections attempt to shed some light on these observations by examining the characteristics of BHCs employing different corporate control mechanisms. Characteristics of firms subject to different control changes. I focus on a number of performance, ownership, and board characteristics of BHCs, on the assumption that these variables may determine which (if any) control devices are used. Definitions and sources for these variables are given in Table 2. Table 1 Frequency of Alternative Corporate Control Changes (Percent of total sample) In MSV’s sample of 454 manufacturing firms Hostile takeover In 234 bank holding companies 8.8 1.7 20.5 10.2 7.5 10.7 36.8 22.6 Regulatory intervention 0 14.1 Total control changes 36.8 36.7 Management turnover Friendly merger Market-based control changes 28 Table 2 Data Definitions and Sources I use two different measures of performance of the BHC under existing management: stock market abnormal returns and a return on equity accounting measure. The stock market measure of performance (RETURN) is the cumulative abnormal return over the period 1985 – 86, calculated using the capital asset pricing model (CAPM) parameterized over the fouryear period 1983 – 86.17 The data for returns are the standard monthly series from the CRSP tapes. This performance measure is calculated over a period prior to 1987 to avoid capturing any effects of the market’s anticipations of future corporate control changes. Doing so means it is more likely that my measure is capturing the market’s expectations of future profitability of the BHC under current management, not the expected premium from a control change. The accounting performance measure (ROE) is the average return on equity from COMPUSTAT over the period 1987 to the date of any control change, or 1992 if there were no control change.18 Since this is an accounting measure of performance, there is no contamination from the market’s expectations about future control changes and thus no need to calculate the measure over a period prior to 1987. Ownership characteristics include the equity holdings of insiders (INSIDE ) and outsiders (OUTSIDE ) on the board of directors in 1987 as a percentage of total outstanding shares. Equity holdings of insiders may proxy for the entrenchment of current management and their financial incentive to accept a friendly offer. Outsider equity holdings proxy for the incentive that outside board members have to perform monitoring duties on current management. Insiders are defined as those members of the board who are also members of current management. Outsiders are defined as those board members who are not insiders and also not employees of firms that may have business dealings with the bank. Outsiders include primarily academics, retirees who are not previous employees of the bank, individuals, and those listed as chairmen of investment groups with their own name.19 In addition, the cumulative shareholdings—as a percentage of outstanding shares—of those shareholders holding greater than 5-percent stakes in the BHC in 1987 are reported as large shareholders’ holdings (LARGE ). The greater a large shareholder’s stakes in the company, the greater his or her incentive to ensure that management is maximizing profits. These data are obtained from 10–Ks, proxies, and other SEC filings. Management characteristics include a FEDERAL RESERVE BANK OF DALLAS Variable Definition RETURN Cumulative abnormal return, 1985–86, from the monthly CAPM, estimated over 1983–86 (SOURCE: CRSP). ROE Annual average return on equity, 1987 to year of control change or, if no control change, to 1992 (SOURCE: COMPUSTAT). INSIDE Equity stakes of insiders (current management team) on the board of directors in 1987 as a percentage of total outstanding shares (SOURCE: SEC filings). OUTSIDE Equity stakes of outsiders on the board in 1987 as a percentage of total outstanding shares (SOURCE: SEC filings). LARGE Combined equity stake of greater than 5-percent shareholders in 1987 as a percentage of total outstanding shares (SOURCE: SEC filings). FF Dummy = 1 if any signer of the annual report is a member of the founding family or of the family of a previous signer of the annual report (SOURCE: annual reports, Who’s Who in American Banking). BOSS Dummy = 1 if only one executive signs the annual report and no other executive holds the title of chairman, CEO, or president (SOURCE: annual reports). SIZE Market value of equity in 1987 in millions of dollars (SOURCE: COMPUSTAT). dummy (FF ) indicating whether any signer of the annual report is from the founding family. Top officer members of the founding family were identified from old annual reports and various editions of Who’s Who in American Banking. Members of the founding family that are part of the top management team may have a special ability to resist challenges to their control even without a substantial ownership stake by virtue of having handpicked the board over a long period of time.20 In addition, following MSV, I record a dummy variable (BOSS ) indicating if only one executive signs the annual report and no other executive holds the title of chairman, chief executive officer, or president of the BHC. The BOSS variable tries to identify top executives who either completely dominate the management of the BHC or have no clear replacement and who therefore may be particularly protected from disciplinary action by the board. This variable is constructed from data from the annual report. Table 3 presents the means of performance measures and ownership and board structure characteristics for five categories of firms in my sample. The first four categories include BHCs that experienced one of the four types of corporate control change: management turnover, hostile takeover, friendly acquisition, and regulatory intervention. The fifth category includes the remaining (“no control change”) BHCs that did not experience any control change. Asterisks indicate the statistical significance of differences in the means of the control change groups relative to 29 ECONOMIC REVIEW THIRD QUARTER 1995 Table 3 Performance, Management, and Ownership Characteristic Means By Control Outcome in 234 Bank Holding Companies Number of BHCs Management turnover Hostile takeover Friendly merger Regulatory intervention No control change 22 4 25 33 150 Performance RETURN ROE –11.5%* 5.3% 9.5%*** 5.1%* 12.2% 13.8%*** –14.3%* .2%*** –1.9% 10.2% Firm size (in millions of dollars) SIZE 630.2 354.1* 438.1* 909.2 717.4 Ownership structure LARGE 38.2%* 15.9% 11.7%* 15.0% OUTSIDE 15.1% 1.8%* 1.0% 1.2% .4%* .9% INSIDE 2.9%* 1.2%** 5.0% 2.6%* 4.4% .11 .04* .15 .26 .23 .17 Management characteristics (zero-one dummies) Family founder on management team (FF ) .09* One-person management team (BOSS ) .10* 0 .25 *, **, and *** indicate means are significantly different from the no-control-change category at the 10-percent, 5-percent, and 1-percent levels, respectively. NOTE: For definitions of variables, see Table 2. tervention. While the motivation for regulatory action makes this result for the regulatory group almost a truism, it is also clear that boards of banks do respond, however weakly, to poor performance. The finding that both the stock market and accounting measures of performance are significantly better at BHCs that undergo a friendly merger than at those undergoing no control change suggests that the motivation for such mergers may not be the expectation of better performance resulting from a change in poor managerial policy. Mergers may, for example, be more motivated by the acquirer’s desire to diversify operations across state lines or capitalize upon another bank’s customer base. In these cases, BHCs may look for potential targets that fit their desire to diversify but that are already performing well and do not require the bidder to engage in the costly process of restructuring the bank’s operations and turning the bank around. Table 3 also suggests that size matters in determining the type of corporate control change. For obvious reasons, it appears easier to acquire smaller BHCs, either through friendly merger or hostile takeover. the no-control-change group. Table 3 indicates that firms experiencing management turnover or regulatory intervention have abnormal stock market returns of –11.5 percent and –14.3 percent, respectively, in the period 1985 – 86, compared with –1.9 percent for firms experiencing no control change. Targets of friendly bids have abnormal returns of +9.5 percent, while targets of hostile bids have abnormal returns of +5.3 percent. Each group’s performance is statistically different from that of the no-control-change group, except for the hostile group.21 The same pattern of performance between corporate control groups is exhibited when the measure of performance is ROE: BHCs in the regulatory and management turnover group show significantly poorer performance than the no-control-change group, whereas BHCs subject to a friendly merger show significantly better performance than the no-change group. Performance in the hostile takeover group is not statistically significantly different from that of the no-control-change group. As expected, performance is relatively poor among those BHCs that ultimately undergo either management turnover or regulatory in- 30 The equity stakes of large shareholders, board insiders, and board outsiders are all lower in those BHCs that undergo regulatory intervention than those that do not experience a control change, consistent with the notion that smaller equity stakes lead to lower incentives to ensure the success of the firm or react to poor performance by changing management or management policies. Equity stakes held by board outsiders are higher and stakes held by board insiders are lower in BHCs that undergo management turnover relative to the no-control-change BHCs. This is consistent with the notion that board insiders in these firms are less entrenched and board outsiders more determined to enact change in response to signs of poor performance. In addition, the finding of higher equity stakes held by insiders in BHCs that were the target of friendly offers relative to no-control-change BHCs is consistent with the notion that insiders with large equity stakes may have financial incentives to acquiesce to merger offers that do not involve their immediate removal. The zero-one dummy variable FF has a mean value of 0.09 for a BHC experiencing a management turnover, versus 0.15 for a BHC experiencing no control change. In other words, a BHC that undergoes a management turnover is about 60 percent as likely to have a member of the founding family in a top management position than a no-control-change BHC. Similarly, no BHC that experienced a hostile takeover had a member of the founding family as a member of top management. Family founders may be more entrenched managers because they typically have higher equity stakes and also have had influence over the selection of the board over a long period of time. Similarly, BHCs that experience a management turnover are about 60 percent as likely to be run by a one-person management team (a BOSS ) as a no-control-change BHC. In contrast, targets of hostile takeovers and friendly mergers are about 1.5 times more likely to be run by one-person management teams than no-change BHCs. BHCs that undergo regulatory intervention are also more likely (about 1.35 times) to be run by a BOSS.22 This evidence suggests that ownership and board structure are important in determining the form of corporate control change. Although the scarcity of hostile takeovers in the sample makes it difficult to identify specific characteristics of BHCs more likely to be subject to a hostile takeover, it is easier to identify distinguishing characteristics of BHCs in the three other corpo- FEDERAL RESERVE BANK OF DALLAS rate control change groups. For example, Table 3 suggests that management teams of those BHCs that own large equity stakes, consist of family founders and/or one-person management teams, and whose outside directors hold relatively small equity stakes may be entrenched enough to avoid internal discipline by their board of directors.23 In addition, those BHCs for which market-based corporate control mechanisms fail to operate and that thus become subject to intervention by regulators clearly exhibit lower ownership concentration by large equity holders and by inside and outside board members. Market-based measures of corporate control may fail in these cases because there is no agent in management, on the board, or among shareholders that has a large enough equity stake to provide adequate incentives to monitor the performance of the BHC and take appropriate action when performance begins to deteriorate. The following section investigates whether these conclusions are robust to multivariate analysis. Multivariate analysis of corporate control changes. I present four-choice logit estimates of the determinants of the form of control change. The four choices are complete management turnover, friendly merger, regulatory intervention, and no control change. I delete the hostile takeover choice from my universe since there are so few of these observations (four) in the sample. Table 4 presents the multinomial logit models for two different specifications using two different measures of performance (RETURN and ROE ), along with measures of inside board ownership (INSIDE ), large shareholder ownership (LARGE ), the natural log of BHC size (LN SIZE ), and whether there was a one-person management team in place (BOSS ).24 In each case, the coefficients on the variables for the nocontrol-change group are normalized to zero. Table 5 presents the implied probabilities from the logits for the specification using ROE as a measure of performance.25 Columns 1 and 2 of Table 4 show that using either return on equity (ROE ) or abnormal stock return (RETURN ) as a measure of performance, relative to the probability of being a no-control-change BHC, the probability of top management turnover is higher when the BHC is not run by a one-person management team, when board insiders hold smaller equity stakes, and when the return on equity is lower. The log odds of a management turnover versus no outcome is not significantly affected by the size of the firm or by the combined equity stakes of all greater than 5-percent shareholders. In terms 31 ECONOMIC REVIEW THIRD QUARTER 1995 Table 4 Multinomial Logit Models of Control Outcomes Management turnover Friendly merger INTERCEPT .05 (.20) .15 (.38) –.11 (.31) –.20 (.38) LN SIZE –.03 (.54) –.04 (.80) –.09* (1.7) –.53* (1.8) BOSS – 6.2* (1.9) Regulatory intervention –1.36 (1.4) –1.63 (1.5) –.07* (1.7) .035* (1.8) .04* (1.8) –.06 (.62) –.06 (.56) –.03 (.19) –.06 (.32) INSIDE –.09** (2.3) –.07* (1.7) .001 (.18) –.001 (.21) –.07*** (3.1) –.04* (1.7) LARGE .004 (1.4) .005 (1.5) –.003 (1.1) –.003 (.88) –.02*** (2.8) –.01* (1.8) ROE –.09*** (3.6) — .004 (1.0) –.02*** (2.7) — RETURN — –.35** (2.6) — — –.04** (2.3) — .01 (.15) *, **, and *** indicate statistical significance at the 10-percent, 5-percent, and 1-percent levels, respectively. NOTE: Coefficients on the regression on no-control-change BHCs are normalized to zero. Absolute values of t-statistics are in parentheses. the size of the firm and decrease with the equity stakes of insiders and large shareholders. As one might expect, the odds of regulatory intervention also increase with poorer performance as measured by ROE or RETURN. Column 3 of Table 5 implies that, of these factors, the strongest effects lie in the extent to which large shareholders and insiders own big stakes in the BHC. Starting at the base case, the probability of regulatory intervention increases from 15.6 percent to 22.5 percent as the equity stake held by large shareholders falls from its median value to the top of its lowest quartile value. The probability of regulatory intervention increases from 15.6 percent to 23.4 percent as the equity stake held by insiders falls from its median to the top of its lowest quartile. of probabilities, column 1 of Table 5 indicates that —starting from a “base case” in which LN SIZE and BOSS are set equal to their mean and INSIDE, LARGE, and ROE are set equal to their medians —when ROE falls to the top of its lowest quartile, the estimated probability of a management turnover rises from 11.7 percent to 16.1 percent.26 The estimated probability drops from 11.7 percent to 7.4 percent in the presence of a BOSS, whereas it rises to 14.5 percent in the absence of a BOSS. Similarly, the estimated probability of a management turnover rises from 11.7 percent to 14.6 percent as the insider equity stake falls from its median to the top of its lowest quartile. Columns 3 and 4 of Table 4 show that the log odds of a friendly acquisition relative to no outcome are significantly negatively related to the size of the BHC but to nothing else. In particular, the existence of a one-person management team, board insider, and large shareholder equity stake, and both measures of bank performance (ROE or RETURN ) have no statistically significant influence on the log odds of a friendly acquisition relative to no control change. Consistent with the earlier evidence from the univariate analysis, columns 5 and 6 of Table 4 show that the log odds of regulatory intervention versus no outcome increase with Conclusions In this article, I explore the effectiveness of various corporate control mechanisms in the banking industry. My analysis suggests that while the market-based mechanisms of corporate control in BHCs appear to operate in the same broad fashion as in manufacturing firms, there may be weaknesses in the effectiveness of two aspects of the corporate control mechanism in BHCs: hostile takeovers and intervention by the board of directors. These weaknesses 32 Table 5 Estimated Probabilities from Multinomial Logit Model* may make the corporate control problem in banking more severe than in nonbank sectors. My analysis confirms the conventional wisdom that hostile takeovers do not play an important role in disciplining management in BHCs. I found little evidence of the disciplinary role of friendly mergers, which appeared to take place primarily among BHCs that were performing well. This result suggests that the main motivation for friendly acquisitions may be for reasons other than disciplining current management to increase shareholder value. If so, the primary responsibility for disciplining managers at BHCs rests with boards of directors. Boards of BHCs (like those of manufacturing firms) do appear to respond to poor performance. Both the univariate and multivariate analysis imply that poor performance increases the probability the board will discipline current management. Overall, however, boards appear to be less assertive in their corporate governance responsibilities than in manufacturing firms. Board-induced turnover of current management in my sample of BHCs is half as frequent as in MSV’s sample of manufacturing firms.27 Why might this be the case? Recall that, like boards of manufacturing firms, bank boards appear weaker in disciplining management when managers are entrenched because of relatively high levels of insider ownership or low levels of board outsider ownership, or when one-person management teams are in place. Thus, management may be more insulated from board action in banks if bank managers hold more equity than do managers at nonbanks, if one-person management teams are more frequent among BHCs than they are among nonbanks, or if outside board member ownership is lower at banks. The evidence suggests that at least the first two factors cannot explain the weakness of bank boards. One-person management teams appear no more frequent among BHCs than among manufacturing firms. In MSV’s sample of manufacturing firms, one-person management teams occur with a frequency of 23.3 percent, while they occur with a frequency of 19.7 percent in my sample of BHCs. Similarly, insider equity stakes do not appear larger in banks than in nonfinancial firms. Byrd and Hickman (1992) report that the mean and median insider equity stakes for their sample of nonfinancial firms are 10.9 percent and 2 percent, respectively, compared with 4.1 percent and 1.3 percent for my sample of BHCs. Outside directors, however, do appear to take larger stakes in nonfinancial firms than in banks, judging by a comparison with the Byrd FEDERAL RESERVE BANK OF DALLAS Probability of Management turnover Friendly merger Regulatory intervention Base case* .117 .095 .156 BOSS present .074 .096 .158 No BOSS present .145 .094 .152 ROE at top of lowest quartile .161 .088 .157 LARGE at top of lowest quartile .110 .095 .225 INSIDE at top of lowest quartile .146 .088 .234 * Base case is from the first specification in Table 4 where LN SIZE and BOSS are estimated at their means for the entire sample, and LARGE, INSIDE, and ROE are at their medians. The rows following the base case are estimated probabilities evaluated at various points, differing from the base case only in the value of the indicated independent variable. and Hickman study. They found the mean and median equity stake held by board outsiders in their sample of firms was 2 percent and 0.08 percent, respectively, compared with 1 percent and 0.05 percent for my sample of BHCs. Thus, boards conceivably may be weaker in banks because outside directors hold less equity and are presumably less motivated to impose disciplinary measures on management. Whatever the reason for weaker boards among BHCs, when combined with the regulatory impediments to hostile takeovers, weaker boards may contribute to a corporate governance mechanism in banks that is not as efficient at disciplining managers as those mechanisms in other sectors. For example, MSV found that corporate boards were particularly weak in removing unresponsive managers in manufacturing firms that were in declining sectors and that required radical downsizing and restructuring. In these sectors, the restructuring function was primarily performed by hostile takeovers. MSV term this situation a third-best solution, on the grounds that internal control devices are inherently cheaper to operate and more conducive to long-term planning than are hostile takeovers. In the banking industry, however, while boards are even weaker than in manufacturing sectors, the use of hostile takeovers as an important method of restructuring is also ruled out. By default, this void has given regulators a primary role in providing a last-resort control mechanism—what might be termed a fourth-best solution since takeover by regulators is almost certainly far more costly than any market-based alternative. 33 ECONOMIC REVIEW THIRD QUARTER 1995 These results suggest that policymakers should take corporate control issues seriously when considering legislative alternatives to the current system of bank regulation and organization. In particular, the finding that banks that have undergone regulatory intervention have markedly lower ownership concentration than other banks suggests that higher ownership concentration among banks might improve performance by motivating greater oversight and monitoring by large stakeholders and their representatives on the board of directors. If so, current restrictions on potential owners of commercial banks may have costs. Some of the proposed banking legislation in Congress could also be evaluated in this light, since different proposals vary quite substantially in the degree to which they relax the current restrictions on permissible bank owners. In addition, the absence of a credible takeover threat among banks appears to have a marked influence on the effectiveness of the corporate control mechanism operating in banks. While regulators have been careful not to discriminate actively against bank mergers on the basis of whether they are hostile or not, the long regulatory process that all bank mergers have to go through tends to make hostile takeovers much more difficult to achieve than friendly mergers. This suggests that there may be beneficial effects on the corporate control mechanism in banks from removing some of the more obvious obstacles to hostile takeovers in banking by, for example, relaxing interstate banking regulations and increasing the speed with which regulators process merger applications. 6 7 8 Notes 1 2 3 4 5 I thank Allen Berger, Mark Carey, Sally Davies, Harvey Rosenblum, Myron Kwast, Tom Siems, and Jim Thomson for comments and useful conversations, and Ed Ettin for suggesting this line of research. I also thank Rebecca Menes for extraordinary diligence in collecting the data and Jim Yeatts for research assistance. For some documentation of these trends, see Gorton and Rosen (1992). Note that the claim that the banking industry is in decline is by no means universally accepted. On this issue, see Boyd and Gertler (1994), Levonian (1995), Kaufman and Mote (1994), and articles in the Federal Reserve Bank of Chicago (1994). See, for example, Ely (1992). See Keeley (1990) and McManus and Rosen (1991). Risk-based deposit insurance premiums were introduced by a provision of the FDIC Improvement Act in 1993. This change does not affect my empirical results since my sample period ends in 1992. See, for example, Edwards (1977), Glassman and 9 34 Rhoades (1980), Hannan and Mavinga (1980), Smirlock and Marshall (1983), James (1984), and Brickley and James (1987). Following MSV, I focus on complete rather than partial turnover of the signers of the annual report over a fiveyear period because I am interested in disciplinary management changes forced by the board. Most of the changes in which one cosigner of the annual report replaces another (partial turnover) likely represent ordinary succession rather than disciplinary action by the board. Of course, counting as disciplinary turnover all cases where the list of signers in 1987 was completely different from the list in 1992 may include some cases where there were two or more ordinary successions (partial turnovers) within the five-year period that resulted in none of the 1987 signers being signers in 1992. This multiple partial turnover phenomenon, in fact, occurs in only two cases in my sample. When making comparisons with the frequencies reported by MSV, I count these two cases as management turnover in order to maintain consistency with MSV’s definition. I do not count these cases as management turnover in the remainder of this article. There are twenty-two when the two multiple partial turnover cases are excluded. Enforcement actions can be formal or informal. Formal actions range from cease and desist orders to civil money penalties on managers and directors. Formal actions are regulators’ most severe forms of action and are always made public by regulators. Informal actions range from commitment letters —which set forth the reforms the BHC needs and the time frame within which those reforms are to be achieved—to memorandums of understanding, a document drafted by regulators and signed by every member of the BHC board. Informal actions are not made public by the regulatory authorities. In some but not every case, informal actions will be disclosed by the BHC itself if it is making a security offering and the enforcement action is deemed to be material information to potential investors. See Rockett (1994). The composite BOPEC rating reflects evaluations on a scale from 1 (strongest) to 5 (weakest) and is arrived at by combining the individual ratings assigned to the BHC in five different component areas (each of which contributes a letter to the acronym BOPEC); namely, the Bank subsidiaries, Other nonbank subsidiaries, the Parent company, the level of consolidated Earnings, and the level of Capital adequacy. As such, the BOPEC rating system for BHCs is structured very much like the CAMEL rating system for individual banks. The decision to impose specific enforcement actions generally depends on the composite BOPEC rating the institution receives in its periodic examination by regulators. If an examination results in a composite BOPEC rating of 3 or below, then the BHC is likely to require “more than normal” supervision by the 10 11 12 13 14 15 16 17 18 19 regulatory authorities (see Federal Reserve Regulatory Service, vol. 2, paragraph 4-865). Except where noted, the results using the BOPEC ratings to construct the regulatory intervention group are qualitatively similar to those presented here. Such were the problems of banks during 1989 –91 that falling in the bottom quintile of the sample may have been sufficient to trigger some regulatory intervention. Again, except where noted, the results using the bottom quintile of the sample as the regulatory intervention group are qualitatively similar to those presented here. 20 21 Of course, this is in part an artifact of my definition of the regulatory intervention category for BHCs as constituting those BHCs that appear in the bottom decile of the sample ranked by the bad loan ratio. Defining the regulatory intervention group as the bottom quintile of firms ranked by this measure, or alternatively, by those BHCs with a BOPEC rating of 3 or below during the sample period, increases the number of BHCs in the regulatory intervention group substantially. Of course, comparing frequencies of total corporate changes assumes that firms in the two samples are subject to the same degree of corporate control problems ex ante the use of corporate control mechanisms considered in the article. In other words, that management is being disciplined to the same extent by other corporate control mechanisms not considered here, such as pay-for-performance compensation packages and competition in product markets. On this point, Houston and James (1993) present evidence that the sensitivity of CEO pay to firm performance is significantly lower in banks than among nonbanks. This finding, combined with the traditional partial insulation from competition in product markets that banks enjoy owing to their ability to issue insured liabilities, suggests that the need for the corporate control mechanisms considered in this article may be greater in banking than in other industries. In fact, as mentioned earlier, alternative plausible definitions of the regulatory intervention group yield a much larger number of BHCs in this group. My measure of turnover here includes the two previously noted cases of multiple partial turnover in order to maintain consistency with the definition used by MSV. Houston and James (1993) use a different measure of management turnover and find that management turnover in banks is somewhat less than in a sample of nonbanks but that the differences are not statistically significant. I restrict myself to the period 1983–86 to parameterize the CAPM because Kane and Unal (1988) identify a break in the return-generating process for banks in 1982 related to changes in the regulatory and financial environment of banks during that year. ROE is defined as income before extraordinary items divided by common equity. FEDERAL RESERVE BANK OF DALLAS 22 23 24 25 26 27 This follows Hermalin and Weisbach (1988) and Byrd and Hickman (1992), who define an outsider more narrowly than just those who are not insiders. For this reason, I set FF = 1 for those BHCs for which a signer of the annual report was related to an immediate previous signer of the annual report, regardless of whether the signer was a member of the founding family. Since the hostile takeover group consists of only four BHCs, it is hard to get statistically significant differences between it and the no-control-change group in all but a few variables. Nevertheless, the higher abnormal return posted for this group may reflect some contamination from investors’ expectations of a future control change. Note, however, that these last two differences are not statistically significant. These are essentially the conclusions of MSV from their analysis of a sample of manufacturing firms. A number of other specifications were tried. The family founder dummy (FF ) showed the same sign and significance pattern as the INSIDE variable when used in the specification in place of INSIDE. When included together with the INSIDE variable, FF became insignificant. The implied probabilities for the alternative measure of performance — abnormal returns — were little different from those presented here. I must start from a set of initial conditions — a “base” case — since the marginal effects of the regressors upon the implied probabilities in a multinomial logit model depend upon the initial values of all the independent variables. See Maddala (1983). One manifestation of this weakness may be in the fact that boards of BHCs are about 50 percent larger than boards of nonfinancial firms. The mean number of directors in my sample of BHCs is 18, compared with 12.1 for Byrd and Hickman’s (1992) sample of nonfinancial firms. Large boards are likely more unwieldy and less capable of responding quickly to management problems. If members of management realize this, then they may seek to entrench themselves by increasing the size of the board. References Allen, Linda, and A. Sinan Cebenoyan (1991), “Bank Acquisitions and Ownership Structure: Theory and Evidence,” Journal of Banking and Finance 15 (April): 425 – 48. Baradwaj, Babu G., Donald R. Fraser, and Eugene P. H. Furtado (1990), “Hostile Bank Takeover Offers: Analysis and Implications,” Journal of Banking and Finance 14 (December): 1229 –1242. Boyd, J. H., and Mark Gertler (1994), “Are Banks Dead? Or Are the Reports Greatly Exaggerated?” Federal Reserve Bank of Chicago Bank Structure and Competition: The Declining Role of Banking? proceedings of a conference, 85 –117. 35 ECONOMIC REVIEW THIRD QUARTER 1995 Brickley, J. A., and C. M. James (1987), “The Takeover Market, Corporate Board Composition, and Ownership Structure: The Case of Banking,” Journal of Law and Economics 30 (April): 161– 80. James, C. M. (1984), “An Analysis of the Effect of State Acquisition Laws on Managerial Efficiency: The Case of Bank Holding Company Acquisitions,” Journal of Law and Economics 27 (April): 211– 26. Byrd, John W., and Kent A. Hickman (1992), “Do Outside Directors Monitor Managers?” Journal of Financial Economics 32 (October): 195 –221. Jensen, Michael C., and Kevin J. Murphy (1990), “Performance Pay and Top-Management Incentives,” Journal of Political Economy 98 (April): 225 – 64. Edwards, Franklin R. (1977), “Managerial Objectives in Regulated Industries: Expense Preference Behavior in Banking,” Journal of Political Economy 85 (February): 147– 62. Kane, Edward, and Haluk Unal (1988), “Change in Market Assessments of Deposit-Institution Riskiness,” Journal of Financial Services Research 1 (June): 207– 30. Kaufman, G., and L. Mote (1994), “Is Banking a Declining Industry?: A Historical Perspective,” Economic Perspectives 18 (May): 2 – 21. Ely, B. (1992), “Commercial Banks Are Not Obsolete and the Federal Government Should Stop Trying to Make Them So,” Federal Reserve Bank of Chicago Bank Structure and Competition: Credit Markets in Transition, proceedings of a conference, 356 –90. Keeley, Michael (1990), “Deposit Insurance, Risk, and Market Power in Banking,” American Economic Review 80 (December): 1183 –1200. Federal Reserve Bank of Chicago (1994), Bank Structure and Competition: The Declining Role of Banking? (proceedings of a conference). Levonian, M. (1995), “Why Banking Isn’t Declining,” Federal Reserve Bank of San Francisco Weekly Letter no. 95 – 05. Glassman, Cynthia A., and Stephen A. Rhoades (1980), “Owner vs. Manager Control Effects on Bank Performance,” Review of Economics and Statistics 62 (May): 263 –70. Maddala, G. S. (1983), Limited Dependent and Qualitative Variables in Econometrics (Cambridge, England: Cambridge University Press). Gorton, Gary, and Richard Rosen (1992), “Corporate Control, Portfolio Choice and the Decline of Banking,” NBER Working Paper, no. 4247 (Cambridge, Mass.: National Bureau of Economic Research, December). McManus, D., and R. Rosen (1991), “Risk and Capitalization in Banking,” Federal Reserve Bank of Chicago Bank Structure and Competition: Rebuilding Banking, proceedings of a conference, 296 – 321. Hannan, Timothy H., and Ferdinand Mavinga (1980), “Expense Preference and Managerial Control: The Case of the Banking Firm,” Bell Journal of Economics 11 (Autumn): 671– 82. Morck, Randall, Andrei Shleifer, and Robert Vishny (1989), “Alternative Mechanisms for Corporate Control,” American Economic Review 79 (September): 842– 52. Rockett, J. M. (1994), “Understanding Regulatory Enforcement Actions,” The Bankers Magazine, January. Hermalin, Benjamin E., and Michael S. Weisbach (1988), “The Determinants of Board Composition,” RAND Journal of Economics 19 (Winter): 589 – 606. Smirlock, M., and W. Marshall (1983), “Monopoly Power and Expense Preference Behavior: Theory and Empirical Evidence to the Contrary,” Bell Journal of Economics 14 (Spring): 166–78. Houston, J., and C. M. James (1993), “An Analysis of the Determinants of Managerial Compensation in Banking,” unpublished manuscript, University of Florida. 36