The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
F E D E R A L R E S E R V E B A N K O F AT L A N TA Social Security Private Accounts: A Risky Proposition? GERALD P. DWYER JR. The author is the vice president in charge of the financial section of the Atlanta Fed’s research department. He thanks Lee Cohen, Eric Engen, Linda Mundy, and Paula Tkac for helpful comments and Lee Cohen for research assistance. rivate accounts have been mentioned as one part of the resolution of the difficulties facing Social Security over the coming years. The trustees of the Social Security Trust Funds project that Social Security will become insolvent in 2041 (Board of Trustees 2005). Because Social Security is a government program, the meaning of solvency here is tricky—after all, the federal government is not projected to be insolvent in any meaning of the word. Isn’t 2041 a long time from now? Why not wait for a while before making a change to such an important government program? The purpose of this article is to provide an introduction to Social Security private accounts—what they can and cannot do for those who choose to use them and for looming Social Security deficits. The discussion will also explore why it is expedient to introduce these accounts now to deal with a problem that will not become acute for decades. Several different precise proposals about private accounts for Social Security have been put forward. These proposals vary in terms of the amounts that can be contributed, permissible investments, and the way funds can be used. The article discusses a general version of these private accounts and their implications for private individuals and points out the implications for those who choose them. P How Would Private Accounts Operate? Basic principles of private accounts. The common element in all proposals for private accounts is that people can take part of their Social Security taxes and place them in accounts in which they have personal ownership rights. In most respects, these accounts would be no different than current individual retirement accounts (IRAs) or defined-contribution retirement plans at many places of employment. While many people in the United States have such accounts, not all do. Roughly 40 percent of all families did not have such accounts in 2001 (Aizcorbe, Kennickell, and Moore 2003). Such accounts would make accumulating assets possible for many families ECONOMIC REVIEW Third Quarter 2005 1 F E D E R A L R E S E R V E B A N K O F AT L A N TA that otherwise would not have the opportunity to do so. The funds in a private account would be the property of the person whose payroll deductions contributed to the account. This ownership may sound like a small thing, but it is not. In fact, it has far-reaching consequences. The value of a private account in the future would be determined by the returns earned on the assets in the account, with the owner accruing any gains or losses. In other words, the accounts have an element of risk—a risk that the account It is incorrect to suggest that Social might not have enough funds to pay for Security has no risks. The risks are difplanned expenditures. While the exisferent than those with a private account, tence of risk may seem like a difference between Social Security and private but the risks are real. accounts, it is not. All plans for the future involve risk, and the further out the plans, the greater the risk. Private accounts and Social Security are no exception to that rule. Private accounts just have different risks than Social Security, a point elaborated later. Funds in an IRA can be withdrawn before retirement, but no existing proposal gives participants any right to withdraw funds from a Social Security private account before retirement. Private accounts can be viewed as forced saving—the government forces people to save for their retirement whether they want to or not. Allowing private account owners to withdraw funds before retirement is inconsistent with this forced saving for retirement. Owners of private accounts would not be allowed to withdraw funds, no matter how pressing the possible current use of the funds— whether to finance a vacation, to buy a house, to pay for a child’s education, or to pay medical bills. The funds would be dedicated to financing retirement, even if it is the furthest thing from the owner’s mind. Some proposals require that owners of private accounts use the funds to purchase an annuity at a certain age, further restricting the use of private accounts. In contrast, IRAs and 401(k)s can be taken in a lump sum or they can be annuitized. When an asset is annuitized, the owner of the asset trades it for a set of payments over his or her remaining life. These payments may be constant in dollar terms, or they may increase or decrease over time, but the seller of annuities guarantees a stream of payments in exchange for the asset. The annuity provides these payments over the owner’s remaining life, no matter how short or long that period may be. Requiring annuitization by everyone by a certain age would be disadvantageous to some people, most obviously those who are mortally ill or in poor health, but it benefits those who will live longer and does guarantee that owners of private accounts cannot outlive the payments from their account.1 Such a restriction is an aspect of forced saving for retirement. Just as an IRA account can be willed to heirs at death, any amount in a private account could be given to heirs. The owner owns the funds in the private account, and they can be disposed of as the owner sees fit. The funds can be used to provide for a spouse or children or can be given to charity. Social Security. Private accounts are different from Social Security in several ways. First, the Supreme Court has ruled that participants in Social Security do not have ownership rights such as those held by a purchaser of an annuity or a participant in a retirement plan.2 Congress can change benefits, and the change, as long as it is otherwise legal, is not restricted by benefits previously provided in the law. If Congress were to lower Social Security payments by 20 percent today, no one would have recourse through the courts. People’s recourse would be through the ballot box. 2 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA In fact, Social Security benefits have been reduced in the past (Social Security Administration 2005b, 2005c). The benefit formula was scaled back in 1977. Legislative changes in 1983 made Social Security benefits taxable for some recipients and are gradually increasing the age for full-benefits eligibility from sixty-five to sixty-seven.3 In 1993 the taxable part of Social Security benefits was increased from 50 percent to as much as 85 percent of benefits depending on the recipient’s other income. Social Security Administration statements sent out in 2005, which estimate future benefits, accurately state that Estimated benefits are based on current law. Congress has made changes in the law in the past and can do so at any time. The law governing benefit amounts may change because, by 2042, the payroll taxes collected will be enough to pay only about 73 percent of scheduled benefits. Just as with the hypothetical 20 percent decrease in benefits today, this 27 percent decrease in benefits in 2042 is extremely unlikely to happen as a large decrease in benefits in one year. Changes to prevent this large benefit decrease in one year will happen. What type of changes will occur and whether they will happen in 2005 or 2042 or sometime in between are unpredictable. Hence, it is incorrect to suggest that Social Security has no risks. The risks are different than those with a private account, but the risks are real. Because Social Security retirement payments affect so many people, it is natural for discussions of Social Security reforms to focus chiefly on retirement benefits. (In this article, “Social Security,” unless otherwise qualified, refers to this Old Age and Survivors Insurance benefit.) The Social Security Administration, though, actually has a set of programs including disability insurance. Medicare is another very large government program, which provides medical care for those covered by the program.4 Any person paying into the Social Security system runs the risk of never receiving any benefits. If a person lives until retirement age, he or she can begin collecting Social Security retirement benefits, and if that person dies while receiving retirement payments, Social Security generally pays a continuation of retirement benefits to a spouse. Social Security will also pay benefits to surviving minor children until they are eighteen. Some survivors are also eligible for a death benefit—currently $255. But for a person who passes away before reaching retirement age without a spouse or minor children, Social Security benefits received are zero. Risk and Return: Investments of Private Account Balances Most proposals for private accounts limit the range of assets that can be held in a private account but permit owners to determine their investments based on their own 1. This restriction solves a so-called adverse selection problem that some commentators worry about. 2. This ruling was made in Flemming v. Nestor in 1960. (Some background and the decision are provided by the Social Security Administration 2005a.) In this particular case, a foreigner who was a resident in the United States and paid Social Security taxes for nineteen years was deported because he was a Communist. Congress changed the law to deny Social Security benefits to such people who were otherwise eligible for Social Security. Mr. Flemming sued and lost on his claim for Social Security benefits. 3. The Treasury transfers the receipts of the income tax on Social Security benefits to the Social Security Administration, which makes this tax quite effectively a reduction in benefits for those with sources of income besides Social Security. 4. This program has its own set of problems, which are not discussed in this article. ECONOMIC REVIEW Third Quarter 2005 3 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 1 The Variability of Portfolio Returns with the Number of Stocks in the Portfolio, 1955–2004 Standard deviation of return (percent per year) 6.5 5.5 4.5 3.5 1 3 5 7 9 11 13 15 17 19 21 Number of stocks included in portfolio Note: For the 1955–2004 period, the variability of the return on one stock is the variability of the return on the first common stock with A as the first letter in its name, and the variability of the return on two stocks is the return on a portfolio of the A stock plus an equal dollar holding of the first common stock with B as the first letter of its name; for the portfolio of three stocks, the first stock with C as the first letter of its name is added to the portfolio, and similarly for D, E, and so on. Because there is no stock whose name starts with Q, X, Y, or Z for 1955 to 2004, the largest portfolio in the figure includes twenty-two stocks instead of twenty-six. Source: Center for Research in Security Prices (CRSP) (2005) preferences about risk. Generally speaking, the more risk associated with an asset, the higher the expected return. Some people have very low tolerance for risk, but others are quite willing to give up some predictability of the asset’s ultimate payoff in exchange for a higher expected payoff. A typical list of assets that can be held in a private account is similar to the assets held in the Thrift Savings Plan operated by the federal government for civilian employees and members of the armed forces and in many similar private plans. Participants in the Thrift Savings Plan can hold their assets in a government securities investment fund, a fixed income index investment fund, a common stock index investment fund, a small capitalization stock index investment fund, and an international stock index investment fund.5 How do such assets differ, and what factors are likely to determine people’s choices of assets? It is important to note that individual stocks are not included in this list of assets. Individual stocks have substantial idiosyncratic risk that is associated with the individual firm’s fortunes and largely unrelated to other developments. All of these funds are portfolios of assets, which diversify away much of the risk in individual securities. Precisely because idiosyncratic risk is specific to a particular firm, when the value of one firm goes down, another may go up, go down, or stay the same. The idiosyncratic risk of individual firms averages out across firms. Figure 1 shows the variability of the return from holding individual common stocks and how this variability falls as more stocks are held. Variability of return is a common measure of risk. Risk means the probability of a loss, but if bad outcomes are offset by good outcomes, stocks with returns that vary more down also vary more up. Hence, risk can be measured by variability up and down, which the standard devi- 4 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA ation in Figure 1 reflects.6 The stocks shown in Figure 1 were chosen quite arbitrarily; they were picked based on the alphabet and the company’s name.7 As the figure shows, in general, holding more stocks lowers the risk—measured by the variability of the annual return over fifty years—of the holdings of stock. In many ways, this finding just reflects the old adage, “Don’t put all your eggs in one basket.” In the context of stocks, the adage translates into “Don’t put all your retirement in one stock.” Portfolios of a large number of stocks diversify away much of the idiosyncratic risk. If the All plans for the future involve risk, and price of General Motors stock falls today, it will have little effect on a portfolio of the further out the plans, the greater the hundreds of stocks. risk. Private accounts and Social Security Portfolios of assets still have risk. are no exception to that rule. Besides idiosyncratic risk, individual stocks have common risks such as recessionary risk, which remains in a portfolio of stocks. Recessions are associated with decreases in stock prices (Dwyer and Robotti 2004). Individual stocks respond differently to recessions because recessions affect some firms more than others, and holding a portfolio of stocks can reduce the effect of a recession compared to holding a stock highly sensitive to recessions. Nonetheless, no amount of averaging of individual stocks’ recession risk makes that risk go away. Hence, even portfolios of stocks—and portfolios of other assets—have risk. The annual returns on assets such as those in the federal government’s Thrift Savings Plan are shown in Figure 2 for the last fifty years—1955 to 2004. The return for each portfolio is the percentage change of the value of its assets from the end of one year to the end of the next year. The portfolios include Treasury bills (which are short-term government securities), corporate bonds, common stocks, small-cap stocks, and international stocks. Treasury bills, because they have short maturities and are obligations of the federal government, are risk free in the sense that anyone holding them to maturity will get the promised payment of dollars. This return provides a benchmark for measuring the variability of returns. Corporate bonds are longer-term securities that have more risk because they are long term and because they are obligations of companies that may have difficulties and may not make the promised payments to bondholders. The common stock return is similar to that provided by some mutual funds that attempt to produce the same return as an index of overall market returns. The small-cap return shows the higher return, at least over this period, from holding smaller firms’ stocks and the higher variability of the return from holding only such stocks. Even though similar data are available only since 1970, Figure 2 also shows the return on a portfolio of international stocks. These stocks, while not necessarily lower risk than U.S. stocks, can be used to diversify away some of the risk resulting from U.S.-specific events. All of the assets included in Figure 2 are returns on financial assets—pieces of paper that are claims to future payment. These assets do not represent direct 5. Information about the Thrift Savings Plan is from its Web site, www.tsp.gov. 1 6. The standard deviation of a return is s = [Σ(rt – r– )2 / (T – 1)] /2, where rt is the return in any individual year, T is the number of years for which returns are available, and r– is the average return over T years. 7. The choice of stocks available in 1955 is not random because the stocks must exist for the whole period, meaning that the firm cannot have gone out of business or been purchased by another firm. This nonrandom choice does not affect the point about the variability of return as the number of stocks held increases. The returns are from the Center for Research in Security Prices (CRSP) database. ECONOMIC REVIEW Third Quarter 2005 5 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 2 Returns from Portfolios of Financial Assets, Percent per Year, 1955–2004 Treasury bills 70 45 20 –5 –30 Long-term corporate bonds 70 45 20 –5 –30 Common stocks 70 45 Percent 20 –5 –30 Small-cap common stocks 70 45 20 –5 –30 International stocks 70 45 20 –5 –30 Inflation 70 45 20 –5 –30 1950 1960 1970 1980 1990 2000 Source: Ibbotson Associates (2005) ownership of real assets, such as a house owned by a household or a factory owned by a firm. Part of the return from holding a real asset is the capital gain—the increase in price. The capital gain on real assets is positively correlated with the inflation rate for goods and services. The remaining part after inflation, which is called the real return, is the part represented by the value of living in the house or using the factory and represents an asset’s return adjusted for inflation. Financial assets have a related real return: the nominal return adjusted for inflation. Figure 2 includes the inflation rate, which is 6 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA related to the capital gain from holding real Table Risk and Return of F inancial Assets, 1955–2004 assets and can be subtracted from the nominal returns to get an approximate measure Variability Average return of return of the real return from holding these finan(% per year) (% per year)a cial assets. Inflation, while it has varied over the last fifty years, has not varied nearly as Treasury bills 5.28 2.89 much as the returns on these financial Corporate bonds 6.80 10.18 assets other than Treasury bills. Common stocks 10.94 17.66 The variation of annual returns is what Small-cap stocks 14.56 25.64 might be expected. Treasury bills—the International stocks 20.70 56.21 short-term government security—show a The variability of return is measured by the standard deviation of the least variation. The portfolio of corpothe return. rate bonds shows more variation but not Source: Ibbotson Associates (2005) as much as the portfolios of stocks. The stock portfolios show by far the most variation. Small firms have more variable returns than large firms. While diversifying away the idiosyncratic risk of individual securities lowers risk, there still is substantial variation in the returns on stock. The table shows that higher risk is associated with higher average return. The table shows the average return from holding each of these assets and the variability of the return. Treasury bills have the lowest average return as well as the lowest value of the measure of variability of return—the standard deviation. There is a clear relationship between risk and return. In terms of retirement accounts, a different and more informative view of the risk and return from holding an asset is provided by the cumulative value from an investment. One way of seeing this is in terms of a graph showing what would have happened over the past fifty years to an investment of a dollar. Figure 3 shows the cumulative value from investments in these assets other than the international stock and the cumulative value for inflation. The average return in the table for holding stock— 12.3 percent—is twice the average return for Treasury bills—5.3 percent—but the cumulative value for holding stock is ten times higher than for Treasury bills. A dollar invested in Treasury bills without any money taken out for fifty years becomes $12.71 in 2004. That same dollar invested in stock becomes $134.62, more than ten times more. In at least one way, these nominal values are misleading. It is true that $1 put into Treasury bills in 1955 would provide $12.71 in 2004. On the other hand, inflation over this fifty years means that $12.71 in 2004 would not buy nearly as much as $12.71 in 1955. In fact, $12.71 in 2004 would buy only as much as $1.87 in 1955 because prices were 6.8 times higher in 2004 than in 1955. Most of the increase in value resulting from investing in Treasury bills merely reflects inflation. The cumulative value of the investment in stocks also buys less because of inflation, but the $134.62 received after fifty years will buy as much as $19.77 in terms of 1955 prices, far more than the $1.87 from investing in Treasury bills. What may seem like small differences in return become large differences over time. At the end of a year, a dollar invested in Treasury bills that earned the average return would be worth $1.053 while the same dollar invested in a portfolio of stocks would be worth $1.123. At the end of two years, the cumulative values with no money taken out are $1.11 and $1.26, still not much different. At the end of five years, the cumulative values are $1.29 and $1.79. At the end of ten years, the cumulative values are $1.68 and $3.19. The higher average return on stock compounds—the higher return earns a higher return earns a higher return—and what seem like small differences over a short period become large differences over long periods. ECONOMIC REVIEW Third Quarter 2005 7 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 3 Cumulative Values from Portfolios of Financial Assets, 1955–2004 1000.0 Small-cap stocks Logarithmic scale 100.0 Common stocks 10.0 Treasury bills Consumer price index 1.0 Long-term corporate bonds 0.1 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Source: Ibbotson Associates (2005) These cumulative values are based on the average return, which is much like the average number of people in a family. No one gets the average return on these investments every year, any more than any family in the United States has the average number of people—3.13 people in 2003 (U.S. Census Bureau 2004, 52). Stocks in the United States have generated higher cumulative values over time, but there is substantial variability in the return in individual years. That said, over time, those who have invested in stock for a substantial period of time have in fact been rewarded for bearing this risk by higher returns, which result in higher cumulative values. The idiosyncratic variation in particular years becomes less important over longer periods (Siegel 2002). Common advice concerning saving for retirement is that someone far from retirement should invest more in stocks because a few bad years are likely to be balanced with some other good years. This advice appears to be correct because much of a young person’s wealth really is not financial; his wealth is based on future earnings. For a person close to retirement, his or her future earnings are for a shorter period, and a big decrease in stock prices in any one year can have a bigger effect on his or her retirement income (Jagannathan and Kocherlakota 1996). This common advice is reflected in some current mutual funds—called life cycle funds—in which more assets are invested in bonds and less in stocks as a person gets older. Bonds have less variable returns than stocks, and dramatic declines in value are less likely in any individual year. On the other hand, it is worth noting that only long-term bonds fall below the cumulative value for inflation in Figure 3, which indicates that bonds have their own risks. There is risk everywhere. Social Security Deficits and Debt If private accounts have these risks, what problems are private accounts supposed to solve? Can Social Security have any risk? In short, the answer is “Yes, Social Security has a different kind of risk than financial assets, but it has risk.” Why? 8 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA The program commonly called Social Security is more formally called Old Age, Survivors, and Disability Insurance (OASDI), which is administered by the Social Security Administration. This program pays benefits to people if they have paid in sufficient premiums for a sufficient length of time and they (1) become old, (2) become a widow or widower or are a dependent of someone who passes away, or (3) become disabled. These conditions constitute three separate and quite different programs. While there are issues concerning survivors and disability insurance, these issues are dwarfed by the problems confronting the old-age part of OASDI, which I will call Social Security consistent with common use of the term. Medicare has a separate fund and has its own set of substantial problems.8 If the federal government does not lower Social Security has its own risk spending or raise taxes, it will issue more because there is no doubt that the current debt to the public instead of issuing debt to level of tax rates and benefit rates will change. This inevitability sometimes is the Social Security Administration. pointed out by saying that Social Security will become insolvent in the future. This statement is not technically correct because the federal government is not legally obligated to pay any particular level of Social Security benefits, and, hence, benefits can be reduced without legal consequences such as bankruptcy for the government or anything less drastic. The Supreme Court has ruled that Social Security recipients have no standing to sue the federal government to overturn a congressional law denying Social Security payments based on otherwise constitutional criteria (Social Security Administration 2005a). Given these technicalities, what do people mean when they say that Social Security is insolvent or is expected to become insolvent? Social Security’s revenue in 2004 of $658 billion was 31 percent greater than its expenditure of $502 billion (Board of Trustees 2005, Table II.B1, p. 4). A surplus of $156 billion hardly sounds like an insolvent program. Furthermore, the Social Security Trust Fund’s positive balance at the end of 2004 is $1.687 billion, which doesn’t sound bad either. To get an idea of the real problem, Figure 4 presents revenue and benefits for 1990 to 2080. Prospective Social Security payments under current law are indicated by the line “Payable benefits.” Why does a large decrease in payable benefits occur in 2041? To understand this decrease, it is necessary to understand some details about Social Security’s funding and current legal arrangements. Social Security’s total receipts as a percent of taxable payroll are indicated by the line labeled “Receipts.” Social Security currently is running a surplus, taking in more than it pays out. This surplus is indicated in Figure 4 by the greater value of receipts than payable benefits for 2005. This situation is projected to continue until 2017, at which time Social Security will start paying out more than it takes in. This deficit is shown in the figure by the excess of payable benefits over receipts for 2017 to 2041. There is a Social Security Trust Fund, which is the value of accumulated prior surpluses and interest and gives Social Security the right to receive funds from the Treasury. Funds from these accumulated surpluses and interest will be exhausted in 2041, at which time Social Security is required by law to match receipts and expenditures—hence, the big decrease in payable benefits in 2041 and later years. Figure 4 shows the benefits implied by current law but not payable, which continue to increase as a fraction of taxable payroll with the projected aging of the population. 8. These problems are not dwarfed by the old-age part of OASDI (Board of Trustees 2005). ECONOMIC REVIEW Third Quarter 2005 9 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 4 Social Security Receipts and Benefits as a Percent of Taxable Payroll, 1990–2080 25 Scheduled but not payable benefits 20 Payable benefits Percent 15 Receipts 10 5 0 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080 Source: Social Security Administration (2005d) The figure ends with 2080 because Social Security projections extend out only seventyfive years. The graph shows the intermediate forecast from the Social Security trustees (Board of Trustees 2005).9 These figures are based on Social Security collecting taxes under current law and paying benefits under current law. Assumptions about the future can affect the date at which deficits begin and their size, but only extremely optimistic assumptions change the overall pattern in the graph. Social Security is running a surplus now and almost surely will run a deficit in the future. The value of the Social Security Trust Fund is implied by the projected deficits and surpluses each year. When Social Security has a surplus, including interest received on the trust fund, the trust fund increases; when Social Security has a deficit, the trust fund decreases. Figure 5 shows the implied value of the trust fund for 1990 to 2080. The change in the trust fund each year equals the surplus plus interest on the trust fund; when the surplus is positive, the trust fund increases; when the surplus is negative, the trust fund decreases. As Figure 5 shows, the projected trust fund hits zero in 2041 and stays there. This path is very unlikely to happen in exactly this way. Between now and 2041, benefits, taxes, or both are likely to be changed. The federal government could raise taxes to finance the current promised benefits. It could cut benefits so that benefits equal tax revenue under current law. Alternatively, the federal government could lower benefits and raise taxes. The Social Security Trust Fund is held in nonmarketable U.S. government securities. Even though the federal government is running a deficit overall, Social Security is running a surplus. The Treasury provides government securities to the Social Security Trust Fund in exchange for the funds raised by Social Security in excess of those necessary to pay current benefits. This fact has important implications over this seventy-five-year projection. When the trust fund starts to run deficits in 2017, the Social Security Trust Fund will return these nonmarketable securities to the Treasury in exchange for cash. The Treasury, which is not accumulating assets to redeem these securities, will have to raise the cash by either raising taxes, cutting 10 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 5 Trust Fund Balance as a Percent of Taxable Payroll, 1990–2080 3 2 Trust fund balance 1 Percent 0 –1 –2 Balance with scheduled but not payable benefits –3 –4 –5 –6 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080 Source: Social Security Administration (2005d) other spending, issuing more debt, or some combination of the three. If the payments are financed by issuing debt, the debt issuance would be substantial. If the payments are financed by raising taxes or cutting spending, these tax increases or spending cuts would also be substantial. Even if taxes were raised enough in 2041 to finance the scheduled benefits at that time, continuing problems would remain because scheduled benefits continue to increase over the years. Given the existence of this trust fund, it may seem odd that economists refer to Social Security as a “pay-as-you-go” program—a program that uses current taxes to pay current benefits. Are economists perhaps unaware of the trust fund? No. The trust fund is an accounting scheme used by the federal government. The trust fund is similar to a family that has two checkbooks for their single bank account. One checkbook, perhaps savings for college, has a surplus and receives interest. The other checkbook—the main one—is overdrawn for substantially more than the savings for college; this checkbook includes transfers of additional savings and interest to the savings checkbook. Perhaps sometimes it can be useful for a family to segregate accounts to keep track of what is happening, but the real truth of the matter is that, on net, the account is overdrawn. The image of an overdrawn checking account would be overwrought for the federal government’s debt. Still, the net balance is the one that matters for the family’s ability to finance future spending; the same is true for the federal government. In fact, the federal government includes Social Security in the government’s overall—or unified—budget and has done so since 1969. In economic terms, it is the unified budget that matters for the government’s balance sheet and overall state of balance. 9. Projections for a longer period show that any actions that take care of Social Security until 2080 leave a problem in the more distant future. In other words, the program with current tax rates and benefit payments is not sustainable. ECONOMIC REVIEW Third Quarter 2005 11 F E D E R A L R E S E R V E B A N K O F AT L A N TA From the viewpoint of federal government receipts and expenditures, private accounts would be quite different from Social Security. Private accounts would not be the property of the federal government. There would be no justification for including funds put into private accounts in the federal government’s unified budget. This consideration suggests a related question, though. If Social Security no longer receives the funds put into private accounts, won’t this decrease in funds received reduce Social Security’s revenue, raise the government’s debt, and perhaps thereby make Social Security riskier? In part, yes. If private accounts are created, it is hard to imagine including funds put into those private A private account trades the risk of future reductions in Social Security benefits and the accounts in the federal government’s unified budget. Currently, the federal governrisk of dying before retirement for the risks ment is creating nonmarketable bond issues that are given to Social Security in associated with holding financial assets. exchange for the current Social Security surpluses. If the federal government does not lower spending or raise taxes, it will issue more debt to the public instead of issuing debt to the Social Security Administration. In another sense, the answer is no. While not a legal obligation, the Social Security Administration’s revenue can be viewed as having promised future benefits associated with them. A reduction in its revenue that is associated with sufficiently large reductions in promised future benefits can reduce the long-term imbalance between Social Security revenues and benefits. How large do the reductions in benefits have to be for there to be no net effect? The reductions in benefits in exchange for current revenue merely have to reflect the initial amount put into a private account plus interest earned on the nonmarketable government debt held by the Social Security Administration. Most proposals for private accounts reduce future benefits by at least that much. In short, the creation of private accounts need not exacerbate Social Security’s future problems. In fact, creation of private accounts can reduce Social Security’s future problems if the reduction in future benefits is large enough relative to the value of private accounts. The reduction in benefits need reflect only slightly more than the nonmarketable debt no longer created by the federal government and its interest. This feature can be attractive to some people because some people are likely to value a deposit in a private account more than the same funds in Social Security. Is a person better off with a private account or with the possible benefits from Social Security? It depends. Some would find a private account preferable; some would not. Private accounts would tend to be more valuable for those who would prefer receiving the return on financial assets to the likely implicit return on Social Security taxes. In part, one risk is being traded for another. A private account trades the risk of future reductions in Social Security benefits and the risk of dying before retirement for the risks associated with holding financial assets. To the extent that the return on financial assets can be more than the implicit return on Social Security, private accounts can be more worthwhile for those with a longer time until retirement because any difference in return can compound over a longer period. Conclusion There is no such thing as a riskless investment. Private accounts for retirement are not riskless. The closest thing to a riskless investment is a Treasury bill, which can be held to maturity. Even for this asset, returns go up and down over the years with 12 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA interest rates, and inflation reduces the value of dollars received. Stock prices and similar assets have capital gains and losses that can be substantial. On the other hand, the higher risk of loss from holding stocks is associated with higher average returns. Cumulative values from holding stocks over long periods, such as those involved in saving for retirement, are substantially more than cumulative values from holding Treasury bills. Social Security is not riskless either. Without some change in the law, Social Security will not pay out current scheduled benefits for the foreseeable future. Social Security benefits, taxes, or both almost surely will change. What changes will occur is uncertain, so there is risk of a reduction in benefits. This risk is not imaginary; reductions in benefits have happened in the past. Private accounts would provide people with some assets that would diversify their retirement plans away from reliance on scheduled payments from Social Security, especially for those with relatively little in the way of likely retirement income other than Social Security. Private accounts also would lessen the commingling of the government budget and households’ savings for retirement. The decrease in revenue in the unified budget need not mean that government debt will increase when the current scheduled benefits to those paying Social Security taxes are included in government indebtedness. REFERENCES Aizcorbe, Ana M., Arthur B. Kennickell, and Kevin B. Moore. 2003. Recent changes in U.S. family finances: Evidence from the 1998 and 2001 Survey of Consumer Finances. Federal Reserve Bulletin (January): 1–32. Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds. 2005. The 2005 annual report of the Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds. Washington, D.C.: U.S. Government Printing Office. Dwyer, Gerald P., Jr., and Cesare Robotti. 2004. The news in financial asset returns. Federal Reserve Bank of Atlanta Economic Review 89, no.1:1–23. Ibbotson Associates. 2005. Stocks, bonds, bills, and inflation: 2005 yearbook. Chicago: Ibbotson Associates. Jagannathan, Ravi, and Naryana R. Kocherlakota. 1996. Why should older people invest less in stocks than younger people? Federal Reserve Bank of Minneapolis Quarterly Review 20, no. 3:11–23. Siegel, Jeremy J. 2002. Stocks for the long run. 3rd ed. New York: McGraw-Hill. Social Security Administration. 2005a. Supreme Court case: Flemming vs. Nestor. <www.ssa.gov/history/ nestor.html> (May 3, 2005). ———. 2005b. History of the provisions of Old-Age, Survivors, and Disability Insurance. <www.ssa.gov/ OACT/HOP/hopi.htm> (May 3, 2005). ———. 2005c. Research note #16: Summary of major benefits under the Social Security Program. <www.ssa. gov/history/benefittypes.html> (May 3, 2005). ———. 2005d. 2005 OASDI trustees report. <www.ssa. gov/OACT/TR/TR05/trTOC.html> (May 3, 2005). U.S. Census Bureau. 2004. Statistical abstract of the United States: 2004–2005. 124th ed. Washington, D.C.: U.S. Government Printing Office. ECONOMIC REVIEW Third Quarter 2005 13 F E D E R A L R E S E R V E B A N K O F AT L A N TA Buy Foreign While You Can: The Cheap Dollar and Exchange Rate Pass-Through EDUARDO J.J. GANAPOLSKY AND DIEGO VILÁN Ganapolsky is a research economist and assistant policy adviser and Vilán is a senior economic analyst in the regional section of the Atlanta Fed’s research department. They thank Tom Cunningham and John Robertson for their comments and Kelley Heinsman for research assistance. uring 2004, even though the dollar depreciated against several major trading partners’ currencies, the U.S. trade deficit increased, fueled mainly by the high level of imports. Basic economic intuition would tell us that a cheaper dollar would make U.S. imports more expensive and that Americans should thus import less, but it seems that a cheaper dollar did not lead to proportionately more expensive imports. This article presents evidence on the degree of exchange rate pass-through (ERPT) for a wide variety of import categories using monthly data for the period December 1993–December 2004.1 To provide a broad picture of the incidence of the ERPT phenomenon, the analysis first decomposes domestic import prices to their foreign price and exchange rate components. Some econometric exercises then test for the presence of ERPT in selected import categories. These categories are different from those generally used in other studies in many ways, but perhaps most importantly in their level of disaggregation. In general, the data show low ERPT at the monthly frequency over the last decade. The ERPT elasticity of total imports’ prices averages 18 percent—that is, for every 1 percent the dollar depreciates (appreciates), the price of imports increases (decreases) 0.18 percent although this average varies considerably across categories. Items defined as capital goods or consumer goods consistently have low ERPT. On the other hand, most of the results suggest that the dollar’s value does not affect the prices of products in the industrial supplies and materials category. Like previous studies, this study finds a generalized downward trend in ERPT elasticities for the main import categories. At a more disaggregated level, however, the analysis finds several instances of a reversion toward higher ERPT during the last months of 2004. The article begins with a brief review of the empirical literature and a simple decomposition of the import prices data. The presentation of the theoretical model D ECONOMIC REVIEW Third Quarter 2005 15 F E D E R A L R E S E R V E B A N K O F AT L A N TA used and its empirical counterpart emphasizes how to interpret the regressions’ output. The article then describes the data used in the estimations, analyzes the empirical results, and summarizes the main results. A Review of the Literature The economic literature generally supports the partial ERPT hypothesis that only a portion of exchange rate movements will translate into import price changes. Goldberg and Knetter (1997), who provide a comprehensive treatment of the issue, report that previous studies had found lower ERPT in the United States than in other countries. In this respect, they point out that the size of the destination market appears to be important. More recently, Campa and Goldberg (2002) provide cross-country and timeseries evidence for a group of twenty-five Organisation for Economic Co-operation and Development member countries during the 1975–99 period. They also find low pass-through elasticities, both in the short and long run, for the United States. Furthermore, their paper suggests the degree of pass-through has fallen over time, a decline that is explained mainly by the changing composition of the import bundle.2 Olivei (2002) provides estimates of exchange rate pass-through for several import categories for the period 1981–99. The paper reports a substantial degree of variation in ERPT across groups and finds no asymmetric response to appreciations and depreciations. Finally, Marazzi, Sheets, and Vigfusson (2005) find that ERPT to U.S. core import prices declined considerably during the past decade. Apart from previous explanations (a shift toward low pass-through goods in the composition of the import bundle), their study suggests that a geographical reorientation of U.S. imports, a more competitive international market fostered by the presence of China, or the existence of more hedging in the exchange rate markets could explain the phenomenon. Also, the study agrees with the others in that the decline in ERPT seems to be a generalized phenomenon across countries. A Preview of the Facts As mentioned earlier, even though the real exchange rate has been depreciating for some time, the trade deficit has not narrowed accordingly but, on the contrary, has kept increasing. Figure 1 breaks down the trade deficit, imports, and exports into the main categories of traded goods that compose them. The graphs show that the acceleration of the trade deficit’s growth rate is coincident with the rapid increase of deficits in consumer goods and industrial supplies and materials. The acceleration of these deficits is due to rapid growth in imports that is not matched by export growth. Imports of capital goods have also been increasing rapidly, but they have been matched by a prompt increase in their exports. We study import prices at different levels of aggregation, examining the aggregate price index of total imports, the price indexes of the three main import categories (industrial supplies and materials, consumer goods, and capital goods); and, 1. Pass-through can be defined as the percentage change in local currency domestic prices resulting from a 1 percent change in the exchange rate. For the purposes of this study, we focus on the passthrough into domestic import prices. 2. Pass-through elasticities are stable along import categories, but a change toward lower passthrough categories has occurred in the past few years. 16 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 1 The U.S. Trade Deficit, Imports, and Exports 0 Foods, feeds, and beverages Capital goods 5,000 –10,000 0 –20,000 Other goods –5,000 –30,000 Industrial supplies and materials –10,000 Autos –40,000 –15,000 –50,000 –20,000 Consumer goods (nonfood) –25,000 –60,000 Total (left axis) Trade deficit in various categories (in millions of U.S. dollars) Total trade deficit (in millions of U.S. dollars) Trade deficit 10,000 –70,000 –30,000 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 Imports 140,000 40,000 120,000 35,000 Total (left axis) 100,000 30,000 Capital goods 80,000 25,000 Industrial supplies and materials 20,000 60,000 15,000 40,000 Autos 10,000 Other goods Foods, feeds, and beverages Consumer goods (nonfood) 5,000 Imports in various categories (in millions of U.S. dollars) Total imports (in millions of U.S. dollars) 45,000 20,000 0 0 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 Exports 80,000 Capital goods 70,000 30,000 60,000 25,000 Total (left axis) 50,000 20,000 Industrial supplies and materials 40,000 15,000 30,000 Foods, feeds, and beverages 10,000 20,000 Consumer goods (nonfood) Autos 5,000 Exports in various categories (in millions of U.S. dollars) Total exports (in millions of U.S. dollars) 35,000 10,000 Other goods 0 0 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 Note: Consumer goods and capital goods do not include autos. Source: Haver Analytics ECONOMIC REVIEW Third Quarter 2005 17 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 1 Pass-Through and No Pass-Through Frequencies Whole sample Depreciation Appreciation Passthrough No passthrough Passthrough No passthrough Passthrough No passthrough Total industrial supplies and materials Plastic materials Organic chemicals Iron and steel mill products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite and aluminum Lumber Shingles and wallboard 0.371 0.511 0.405 0.435 0.415 0.397 0.435 0.450 0.415 0.450 0.527 0.511 0.629 0.489 0.595 0.565 0.585 0.603 0.565 0.550 0.585 0.550 0.473 0.489 0.362 0.544 0.614 0.579 0.528 0.667 0.596 0.544 0.642 0.561 0.439 0.404 0.638 0.456 0.386 0.421 0.472 0.333 0.404 0.456 0.358 0.439 0.561 0.596 0.378 0.378 0.486 0.500 0.492 0.541 0.514 0.554 0.523 0.527 0.500 0.541 0.622 0.622 0.514 0.500 0.508 0.459 0.486 0.446 0.477 0.473 0.500 0.459 Capital goods except automotive Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Civilian aircraft Medicinal equipment Photo, service machinery 0.667 0.504 0.427 0.878 0.797 0.641 0.626 0.000 0.511 0.473 0.333 0.496 0.573 0.122 0.203 0.359 0.374 1.000 0.489 0.527 0.672 0.544 0.614 0.070 0.170 0.333 0.298 0.800 0.491 0.544 0.328 0.456 0.386 0.930 0.830 0.667 0.702 0.200 0.509 0.456 0.662 0.311 0.419 0.122 0.138 0.216 0.230 0.500 0.311 0.392 0.338 0.689 0.581 0.878 0.862 0.784 0.770 0.500 0.689 0.608 Consumer goods Apparel, household goods–cotton Furniture, household goods Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles–non-wool or cotton 0.417 0.420 0.382 0.511 0.420 0.703 0.117 0.496 0.359 0.500 0.521 0.458 0.583 0.580 0.618 0.489 0.580 0.297 0.883 0.504 0.641 0.500 0.479 0.542 0.448 0.316 0.509 0.386 0.316 0.094 0.295 0.351 0.526 0.528 0.455 0.340 0.552 0.684 0.491 0.614 0.684 0.906 0.705 0.649 0.474 0.472 0.545 0.660 0.392 0.419 0.500 0.338 0.351 0.200 0.280 0.297 0.473 0.323 0.420 0.338 0.608 0.581 0.500 0.662 0.649 0.800 0.720 0.703 0.527 0.677 0.580 0.662 Note: The sample period is December 1993 to December 2004. Frequencies represent the ratio between the number of times a particular event occurred and the total number of events. Source: Authors’ calculations based on data from the BLS; category names are based on BEA end-use categories. finally, at the most disaggregated level, the price indexes of the items that make up to two-thirds of each category. Table 1 reports for each item the frequency with which the monthly changes of the exchange rate and the domestic price move in the same or different directions, defining these events as “pass-through” or “no pass-through.” The frequencies are computed using the import price indexes published by the U.S. Bureau of Labor 18 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Statistics (BLS) and the inverse of the broad nominal dollar index published by the Board of Governors of the Federal Reserve System.3 Then we identify the items for which pass-through or no pass-through constitutes the bulk of the cases, setting twothirds as our threshold.4 This exercise is performed for the whole sample, and the sample is also divided between depreciations and appreciations to determine whether any sign of asymmetric ERPT occurs. The results do not show strong evidence in favor of either the pass-through or the no pass-through hypothesis. For the entire sample, we find clear evidence of no pass-through for just two items in the conWhile industrial supplies and materials sumer goods category (apparel/household goods–cotton and gems, diamonds). show low correlation between foreign For the split sample, we seem to prices and exchange rates, capital goods uncover different behaviors of some prices and consumer goods show highly negative during depreciation and appreciation events. For example, within the industrial correlation coefficients. supplies and materials category, one item (crude oil) shows evidence of pass-through when the dollar depreciates but no evidence of price reduction when the dollar appreciates. The capital goods category presents some interesting observations. Four items (computer accessories, computers, semiconductors, and telecom equipment) demonstrate no pass-through during depreciations but show pass-through during appreciations. Finally, within the consumer goods category three items (toys, games, sporting goods; TVs, VCRs, etc.; and apparel, textiles–non-wool or cotton) do not pass through when the dollar depreciates. Figure 2 shows the decomposition of the monthly change of the dollar price of imported goods into its two components: (1) the change in the foreign currency price of the goods and (2) the change in the dollar price of foreign currencies. To construct this figure, we computed the monthly change of the domestic price and exchange rate indexes and obtained the monthly change of the goods’ foreign currency price as a residual by purging the exchange rate variation from the domestic import price. The figure suggests that most of the changes in the aggregate import price index are driven by the industrial supplies and materials import index while consumer goods and capital goods import prices remain quite flat. Decomposing those variations shows that in the consumer goods and capital goods cases, dollar depreciations (appreciations) are matched fairly closely by reductions (increases) in the foreign price, and therefore the dollar price of these categories shows little ERPT. On the other hand, the foreign price of industrial supplies and materials seems more volatile and less related to changes in the nominal exchange rate. In other words, the volatility of foreign prices is wiping out any possibility of ERPT. These observations are very important to interpreting the potential sources of a low ERPT coefficient. On one hand, it could be the result of a highly negative correlation between nominal exchange rates and foreign prices. On the other hand, it could result from the combination of a very low correlation between nominal exchange 3. This index is expressed in the amount of foreign currency per unit of dollar; we inverted it to measure dollars per unit of foreign currency. Thus, dollar depreciation (appreciation) is a positive (negative) change in the nominal exchange rate index. 4. This test for the ERPT hypothesis is not very stringent given that we define as pass-through any movement in the same direction, independent of the magnitude, of both exchange rates and domestic import prices. As a result, we put the full and partial ERPT concepts together under the pass-through definition. ECONOMIC REVIEW Third Quarter 2005 19 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 2 Import Price Decomposition All categories Industrial supplies .10 .04 Foreign price .05 Percent Percent .02 0 0 –.05 –.02 Domestic price Broad foreign exchange rate index –.04 –.10 –.15 1994 1996 1998 2000 2002 2004 1994 1996 1998 2000 2002 2004 2002 2004 Consumer goods Capital goods .04 .04 .02 Percent Percent .02 0 0 –.02 –.02 –.04 –.04 1994 1996 1998 2000 2002 2004 1994 1996 1998 2000 rates and foreign prices, with a much larger volatility of the latter. However, only in the first case could low ERPT be interpreted as the outcome of foreign firms adjusting markups in response to exchange rate variations. To identify the sources of low ERPT, Table 2 computes the correlations between domestic import price changes and nominal exchange rate changes as well as the correlations between foreign import price changes and nominal exchange rate changes. The results are broadly consistent with those derived from Figure 2 for more aggregated data. In general, evidence supports the partial ERPT hypothesis; the correlation between domestic prices and exchange rates tends to be low. However, when we try to rationalize the sources of the low degree of ERPT, we detect some differences across categories. While industrial supplies and materials show low correlation between foreign prices and exchange rates, capital goods and consumer goods show highly negative correlation coefficients. Indeed, this pattern also holds true at the more disaggregated level. Within industrial supplies and materials, all but three items show low correlation; within capital goods, all but one item show highly negative correlation; and finally, within consumer goods, all the items show strong and negative correlation. Those facts could be interpreted as favoring the explanation of variable markups in the consumer goods and capital goods cases. Interestingly, the buffering effect of markups seems to unwind for industrial supplies and materials, where foreign prices move independently from exchange rates. We complement those observations with the results of Granger causality tests, reported in Table 3.5 In general, we find causality in the Granger sense from exchange rates to domestic import prices for capital goods and consumer goods, but we failed to find any causal relationship for industrial supplies and materials. 20 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 2 Import Price and Nominal Exchange Rate Correlations Depreciation Domestic Foreign price price Appreciation Domestic Foreign price price Domestic price Foreign price Total industrial supplies and materials Plastic materials Organic chemicals Iron and steel mill products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite and aluminum Lumber Shingles and wallboard 0.130 0.184 –0.038 0.123 0.138 0.107 0.003 0.054 0.223 0.020 –0.052 –0.080 –0.285 –0.622 –0.670 –0.368 –0.751 –0.061 –0.111 –0.125 0.132 –0.465 –0.343 –0.448 0.011 0.171 –0.194 0.112 0.038 –0.009 0.012 0.045 0.077 0.019 –0.145 –0.194 –0.255 –0.449 –0.610 –0.310 –0.655 –0.115 –0.074 –0.066 0.011 –0.353 –0.336 –0.436 0.203 0.064 –0.172 0.077 0.174 0.270 0.079 0.155 0.044 0.158 0.177 0.225 –0.107 –0.490 –0.629 –0.187 –0.511 0.147 0.014 0.008 –0.019 –0.163 –0.033 –0.058 Total capital goods except automotive Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Medicinal equipment Photo, service machinery 0.244 0.147 0.231 0.042 0.121 0.105 0.003 0.200 0.194 –0.953 –0.817 –0.895 –0.825 –0.460 –0.718 –0.899 –0.913 –0.847 0.119 0.096 0.212 –0.068 0.068 0.124 0.113 0.041 0.251 –0.920 –0.759 –0.766 –0.754 –0.430 –0.616 –0.749 –0.844 –0.682 0.041 –0.097 0.033 –0.166 –0.033 0.181 –0.070 –0.035 0.060 –0.892 –0.688 –0.864 –0.742 –0.332 –0.400 –0.886 –0.869 –0.760 0.168 –0.094 0.228 0.119 –0.090 0.018 0.192 0.124 0.090 0.185 0.138 –0.011 –0.986 –0.951 –0.920 –0.956 –0.956 –0.879 –0.882 –0.954 –0.958 –0.850 –0.928 –0.949 0.163 –0.051 0.102 –0.041 –0.201 –0.061 –0.148 0.327 0.121 0.129 –0.104 0.022 –0.969 –0.909 –0.913 –0.924 –0.931 –0.804 –0.954 –0.907 –0.924 –0.722 –0.910 –0.896 –0.065 –0.002 0.264 0.091 –0.079 0.095 0.139 –0.114 –0.114 –0.011 0.015 0.159 –0.973 –0.876 –0.715 –0.898 –0.887 –0.731 –0.583 –0.902 –0.906 –0.795 –0.805 –0.871 Total consumer goods Apparel, household goods–cotton Furniture, household goods Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles–non-wool or cotton Note: The sample period is December 1993 to December 2004. Source: Authors’ calculations based on data from the BLS and the Board of Governors of the Federal Reserve System; category names are based on BEA end-use categories. A Framework to Estimate Exchange Rate Pass-Through Theoretical grounds. The literature defines ERPT as “the percentage change in local currency import prices resulting from a one percent change in the exchange rate between the exporting and importing countries” (Goldberg and Knetter 1997, 1248). 5. We test whether causality in the Granger sense exists in either direction between changes in the nominal exchange rate and changes in the domestic import prices. The direction of causality that concerns us is the one that goes from exchange rates to domestic import prices. ECONOMIC REVIEW Third Quarter 2005 21 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 3 Granger Causality Tests Granger causality tests E to ... ... to E Number of observations Total imports 0.502 1.490 130 Total industrial supplies and materials Plastic materials Organic chemicals Iron and steel mill products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite and aluminum Lumber Shingles and wallboard 0.617 2.859* 0.900 0.128 2.932*** 0.099 0.818 0.531 0.832 2.126 0.217 0.042 0.404 0.551 4.078 0.014 0.703 0.352 0.402 2.088 1.627 2.094 2.275 3.568** 130 130 130 130 117 130 130 130 117 130 130 130 Total capital goods except automotive Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Medicinal equipment Photo, service machinery 26.125*** 5.450*** 26.495*** 3.686** 0.740 3.418** 0.727 12.825*** 25.204*** 1.098 0.816 0.041 0.266 1.031 4.716*** 0.899 0.451 0.010 130 130 130 130 117 130 130 130 130 Total consumer goods Apparel, household goods–cotton Furniture, household goods Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles–non-wool or cotton 11.683*** 0.397 3.350** 2.220 0.274 0.415 1.193 2.578* 4.575*** 17.639*** 4.897*** 1.124 0.232 2.394* 0.057 4.078 0.376 0.086 0.012 0.425 0.542 0.328 0.776 0.530 130 130 130 130 130 117 93 130 130 117 130 117 Note: The reported values are F-statistics for estimations with two lags. *,**, and *** indicate significance at the 10 percent, 5 percent, and 1 percent levels, respectively. The sample period is December 1993 to December 2004. Source: Authors’ calculations If the law of one price holds, then exchange rate changes will always pass in full to domestic import prices. This result would also be maintained in the aggregate if purchasing power parity holds.6 But if either the law of one price or purchasing power parity (PPP) fails in any of their versions, then the possibility of having partial ERPT arises. 22 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA If P is the price in local currency of the imported goods, E is the nominal exchange rate, and P * is the price in foreign currency of the imported goods (including transportation, distribution, resale costs, etc.), then PPP implies that P = E ⋅ P *. If P * is independent of E, any change in E will fully transmit into P; this rationale is the essence of full ERPT. However, P * might depend on E: P = E ⋅ P * (E), and therefore the change in P for a given change in E will depend on the behavior of P *. We can assume that goods markets are not perfectly competitive and then write P * as being formed by two components, a markup and the marginal cost of producing (and delivering) the good. Thus, we should reformulate the previous statement: If the markup and the marginal cost of the exporter/producer are both independent of E, then exchange rate movements would fully pass through into domestic import prices. Nonetheless, if either of them are related to E, changes in the exchange rate would imply that ERPT is partial. Evidence in the literature, both at theoretical and empirical levels, indicates that markups and marginal costs would in fact depend on E. Using imperfect competition models, Dornbusch (1987) shows how markup can adjust in response to changes in the exchange rate. Baldwin and Krugman (1989) and Bernard and Jensen (2004) present evidence on the existence of sunk costs to start an export business (advertising, setting up a distribution chain, conducting R&D specific for a market, etc.), which would also help explain markup changes.7 Regarding changes in the marginal costs, according to Feenstra (1989) the exchange rate can enter the cost function directly through the price of imported inputs or indirectly through a change in the scale triggered by the response of demand in the destination market. Burstein, Neves, and Rebelo (2003) show that distribution costs are an important component of retail prices of tradable goods, and, given that distribution activities use nontradables, these could be affected by movements in the exchange rate. 6. In Goldberg and Knetter (1997), the absolute version of the law of one price means that “identical products sell for the same common-currency price in different countries.” On the other hand, the relative version means that “the common-currency prices for a particular product change in the same way in the two countries.” In regard to purchasing power parity, theory requires that the law of one price holds for all the goods in the economy. The absolute version of the law of one price also requires the absence of nontradable goods, and the relative version needs constant nontradable goods prices. 7. A foreign firm would not raise prices or leave the market and allow other firms to enter as soon as it observes the exchange rate falling. Instead, it would absorb the depreciation by reducing its margins. Vice versa, when the exchange rate increases, the foreign firm would revamp its margins without reducing prices in local currency. Obviously, the buffering effect of margins has a limit. On the downside, at some point the foreign firm will decide the effort is no longer productive and will start raising prices. On the upside, when other firms see the thick margins, they will be tempted to sink some resources to enter the market, driving prices and margins down. ECONOMIC REVIEW Third Quarter 2005 23 F E D E R A L R E S E R V E B A N K O F AT L A N TA In sum, we postulate the following import price equation, which is broadly consistent with those behind the empirical exercises in the rest of the literature: (1) P = E ⋅[ψ ( E,.)⋅ c( E,.)], ψ ( E,.) ≡ P *( E,.) , c( E,.) where ψ(.) is the markup that foreign firms charge on their costs and c(.) is their marginal costs. As mentioned earlier, markup depends on market characteristics and demand conditions in the importing country, and, given the relationship of the latter with the value of the local currency, it depends indirectly on exchange rates. The cost of the imported product depends on the price of domestic and foreign inputs and the scale of production, so then it also depends in some way on exchange rates. Empirical counterpart. The empirical implementation of the underlying model in most of the literature follows the regression equation presented in Goldberg and Knetter (1997), which varies from study to study depending on the question the researchers seek to answer and the data they draw on: (2) pt = α + β ⋅ et + δ ⋅ xt + γ ⋅ zt + εt , where all the variables are in logarithmic form, pt is the domestic price of an imported product, et is the nominal exchange rate, xt is a measure of the foreign costs, zt denotes some controls, and εt is an error term. In the general setup, domestic import prices (in local currency) are mainly driven by three variables: (1) the nominal exchange rate, (2) foreign exporters’ costs, and (3) domestic demand (directly through its effect on markup and indirectly through the effects on scale and thus exporters’ costs). Campa and Goldberg (2002) use as proxies for exporters’ costs both an aggregate measure of labor costs in the trading partners and real gross domestic product (GDP) in the domestic country, with the latter trying to capture the effect of demand on the scale and thus on marginal costs. Olivei (2002) combines the nominal exchange rate and foreign exporters’ costs by computing real exchange rates specific for each category of goods. Regarding demand conditions, this study controls for the price of alternative goods with domestic price indexes and for the expenditure on the imported good and its alternatives with U.S. industrial production indexes. Finally, Marazzi, Sheets, and Vigfusson (2005) rely on foreign consumer price indexes (CPI) and producer price indexes (PPI) to capture exporters’ costs and use an index of primary commodities prices to represent the price of alternative goods, which in turn affects domestic demand. The analysis in this article uses the same underlying framework. Like Goldberg and Knetter (1997) and Campa and Goldberg (2002), this article considers nominal exchange rate movements as opposed to real exchange rates.8 On the other hand, we share with Olivei (2002) and Marazzi, Sheets, and Vigfusson (2005) the way we control for foreign costs, using cost proxies specific to each good category, derived either from foreign CPI or PPI; in this article, however, we construct our own indexes. We also share with Olivei the fact that we include U.S. production indexes to control for the state of the business cycle in each sector and use domestic price indexes as proxies for the prices of alternative goods. 24 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA We estimate equation (2) in first differences by using ordinary least squares and recursive least squares methods; specifically, (3) ∆pt = a + b1 ⋅ ∆et + b2 ⋅ ∆ xt + b3 ⋅ ∆zt + vt , where ∆ indicates the first-difference operator, vt is the regression residual, and a and bi are the estimated coefficients. It is apparent from equation (3) that the estimated coefficient b1 is not an estimator of the pass-through elasticity given by β in equation (2). In Appendix A we show that b1 is estimating a quadratic function of the true pass-through elasticity. Therefore, the estimated pass-through elasticities should be computed as the square root of b1. To test for the presence of asymmetries in the pass-through elasticities, we estimate a slightly different version of equation (3): (4) ∆pt = a + b1 ⋅ ∆et + b2 ⋅ ∆xt + b3 ⋅ ∆ zt + b4 ⋅ ∆et Dt + vt , where Dt is a dummy variable that captures the depreciation events.9 In equation (4) we incorporate the interaction term with the aim of testing whether the degree of ERPT is the same or different during depreciation and appreciation events. So in this case b1 estimates some function of the ERPT elasticity when the dollar appreciates, and (b1 + b4) estimates the same function when the dollar depreciates. Thus, our asymmetry test consists of assessing whether b4 is significantly different from zero; if it is, we can reject the hypothesis that ERPT is symmetric. Data Description Import prices and quantities. This article uses monthly import price data from the BLS for the period December 1993–December 2004. The BLS reports price indexes at different levels of aggregation: (1) aggregate import price index (level 1), (2) price index per import category (level 2) (for example, industrial supplies and materials), and (3) price index per item within each import category (level 3) (for example, fuel oil). In this article we work with the three level 2 categories that contribute the most to total imports. Level 3 items are selected so that they explain two-thirds of imports of the corresponding level 2 category. We use annual import data from the U.S. Bureau of Economic Analysis (BEA) for 2002, 2003, and 2004 and from the U.S. Census Bureau for 2001. In some cases a BEA import category does not exactly match the description of a BLS import price category. To reconcile this difference we use our judgment in attempting to find an equivalent category. Table 4 shows all the cases in which the category names from the BEA do not exactly match those from the BLS. Nominal exchange rates. The Board of Governors of the Federal Reserve System constructs three nominal exchange rate indexes: Broad, Major, and Other Important Trading Partners (OITP). The Broad index includes twenty-six currencies from the United States’ main trading partners, the Major index includes the seven 8. Olivei (2002) directly considers the real exchange rate. Marazzi, Sheets, and Vigfusson (2005) consider it indirectly given that they restrict the nominal exchange rate and the foreign price index coefficients to be the same (β = δ). 9. The dummy variable takes the value 1 if the nominal exchange rate depreciates and 0 if it appreciates or remains unchanged. ECONOMIC REVIEW Third Quarter 2005 25 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 4 BEA and BLS Category Matching BEA category BLS category Crude oil Bauxite and aluminum Finished metal shapes Industrial supplies, other Lumber Shingles and wallboard Medical equipment Photo, service machinery Toys, games, sporting goods Household appliances Footwear Writing and art supplies Crude Bauxite, alumina, aluminum and products thereof Finished metal shapes and advanced manufacturing Industrial supplies (aggregate) Lumber and other unfinished building materials Selected building materials Scientific and medical machinery Photo and other service industry machinery Toys, shooting and sporting goods Household and kitchen appliances Footwear of leather, rubber, or other material Other products (notions, writing supplies, tobacco products, etc.) most important currencies, and the remaining nineteen are included in the OITP index. All these indexes are denominated in units of foreign currency per unit of dollar. We use these time series at a monthly frequency. The results reported in this article are based on the Broad index; we also perform some of the exercises with the Major index, but they are robust to this change. Cost proxies. We construct three types of foreign cost proxies for each item and category in the study. The first index is constructed with monthly data from the International Financial Statistics (IFS). Following Campa and Goldberg (2002), we take advantage of the fact that the IFS reports both the real and the nominal exchange rate per country adjusted by labor costs (reu and neu series), and we derive an approximate measure of the trading partners’ costs.10 The other two indexes are both weighted averages of foreign price indexes, yet one is built by combining foreign PPI and wholesale price indexes (WPI) while the other is constructed entirely from foreign CPI. The data we use are monthly. The weights are constructed from the relative importance of each country in the trade volume of each item using the historical monthly import volumes per country from the U.S. Department of Commerce. Industrial production. We use monthly industrial production (IP) indexes constructed by the Board of Governors under the North American Industry Classification System (NAICS). Since both the IP indexes and the trade data from the Commerce Department are built under the NAICS, we must use our judgment to reconcile these variables with the BEA end-use classification system. Table 5 indicates how these categories are matched. In some instances a NAICS category is repeated (for example, computers and computer accessories), and in some others, because no appropriate match is available, we use a category index (a level 2 index). This more aggregate index is able to capture an average of all the changes occurring in a particular sector. We drop out only one item (civilian aircraft) within capital goods because of a lack of sufficient data. Domestic prices. For domestic prices of imported goods, we use two types of indexes: industrial PPI for the items within industrial supplies and materials and capital goods and the urban CPI for all the final goods items within consumer goods. In this case also we must use our judgment when matching the import price items with the categories used as proxies of domestic prices. Table 6 details how all items are matched. 26 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 5 BEA and NAICS Category Matching BEA category NAICS IP category Plastic materials Organic chemicals Iron and steel products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite Lumber Shingles and wallboard Industrial supplies, other Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Civilian aircraft Medicinal equipment Photo, service machinery Apparel, household Furniture, household Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles Plastics material and resin NAICS=325211 Organic chemicals NAICS=32511 Iron and steel products NAICS=3311 Fabricated metal products NAICS=332 Crude oil NAICS=211111 Distillate fuel oil NAICS=32411 Petroleum and coal products NAICS=324 Natural gas NAICS=211111 Alumina and aluminum production and processing NAICS=3313 Wood products NAICS=321 Plywood and misc. wood products NAICS=3212 Level 2 industrial supplies IP index Electrical equipment, appliances, and components NAICS=335 Machinery, except electrical NAICS=33 Computer and peripheral equipment NAICS=3341 Computer and peripheral equipment NAICS=3341 Semiconductor and other electronic components NAICS=3344 Communications equipment NAICS=3342 Aircraft and parts NAICS=336411 Medical equipment and supplies NAICS=3391 Level 2 capital goods IP index Apparel and leather goods NAICS=3152 Household and institutional furniture NAICS=3371 Furniture and related products NAICS=337 Level 2 consumer goods IP index Audio and video equipment NAICS=3343 Level 2 consumer goods IP index Household appliances NAICS=3352 Apparel and leather goods NAICS=3152 Pharmaceutical and medicine NAICS=3254 Paper NAICS=3221 Textiles and products NAICS=313 To test for the presence of unit roots in all the data, we use the augmented DickeyFuller methodology. Because most of the time series in our data set were nonstationary at the 1 percent level of significance, we estimate our models in first differences. Results Table 7 summarizes the results obtained from estimating equation (3). The first column shows the ERPT elasticities obtained from estimating a simple statistical relationship between domestic import prices and exchange rates. The next three columns present the ERPT elasticities estimated using equation (3) for different specifications of the foreign cost: broad, PPI/WPI-based, and CPI-based proxies. The final column gives the estimated ERPT elasticity that we find when using the PPI/WPI-based specification with one more control variable—the domestic price index—which acts as a proxy of the prices of competing goods. 10. The exact derivation and the underlying assumptions are provided in Appendix B. ECONOMIC REVIEW Third Quarter 2005 27 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 6 Import Prices and Domestic PPI/CPI Matching BEA category CPI/PPI category Index Total industrial supplies and materials Plastic materials Organic chemicals Iron and steel mill products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite and aluminum Lumber Shingles and wallboard Industrial supplies, other Intermediate materials: less food and feeds Plastic resins and materials Basic organic chemicals Steel mill products Fabricated structural metal products Crude petroleum Gasoline Petroleum products, refined Natural gas (to pipelines) Primary nonferrous metals (excluding precious) Lumber Building paper and board Intermediate materials: less food and feeds PPI PPI PPI PPI PPI PPI PPI PPI PPI PPI PPI PPI PPI Total capital goods except automotive Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Medicinal equipment Photo, service machinery Capital equipment Electrical industrial apparatus Capital equipment Computer peripheral equipment and parts Electronic computers Capital equipment Telephone and telegraph equipment X-ray and electro medical equipment Capital equipment PPI PPI PPI PPI PPI PPI PPI PPI PPI Total consumer goods Apparel, household goods–cotton Furniture, household goods Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles–non-wool or cotton CPI-U-All Window and floor covering and other linens Furniture and bedding Other household equipment and furnishings Average of sporting goods and toys categories Video and audio Jewelry and watches Household appliances Footwear Medical care commodities Stationery, stationery supplies, gift wrap Apparel CPI CPI CPI CPI CPI CPI CPI CPI CPI CPI CPI CPI The results are robust across different specifications except for industrial supplies and materials, where both the overall category and its related items change substantially when the model is specified with CPI-based proxies.11 We find strong evidence in favor of the partial ERPT hypothesis at the more aggregated levels. In our best (PPI/WPI-based) specification, the total imports category shows an average short-run elasticity of 18 percent for the entire December 1993–December 2004 period. At level 2, the industrial supplies and materials category is more elastic than the level 1 counterpart, but it is statistically significant only in the PPI/WPI-based specification, averaging 29 percent during the sample period. Capital goods and consumer goods are both less elastic than the level 1 counterpart (8 percent and 13 percent, respectively), and, interestingly, they are statistically significant across all specifications. 28 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 7 Pass-Through Elasticities Economic models Foreign costs Statistical model Total imports 0.156* Broadbased PPI/WPI– based CPIbased 0.169** 0.177** 0.170** 0.269 0.215** –0.120 0.201 0.156 0.360 0.220 0.222 0.695** 0.120 –0.144 –0.150 Total industrial supplies and materials Plastic materials Organic chemicals Iron and steel mill products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite and aluminum Lumber Shingles and wallboard 0.241 0.210** –0.106 0.227 0.154*** 0.357 0.121 0.223 0.693** 0.094 –0.168 –0.184 0.263 0.215** –0.105 0.228 0.152* 0.347 0.139 0.226 0.698** 0.129 –0.168 –0.162 0.291* 0.238*** –0.104 0.194 0.161 0.384 0.265 0.262 0.700** 0.100 –0.146 –0.182 Total capital goods except automotive Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Medicinal equipment Photo, service machinery 0.128*** 0.146*** 0.153* 0.075 0.186 0.139 0.026 0.134** 0.158** 0.130*** 0.149* 0.154*** 0.084 0.182 0.138 –0.008 0.131** 0.169*** 0.136*** 0.149* 0.153*** 0.082 0.153 0.141 0.057 0.131** 0.170*** 0.134*** 0.145* 0.148*** 0.089 0.164 0.138 0.049 0.133** 0.174*** 0.084** –0.082 0.143*** 0.090 –0.077 0.026 0.143* 0.092 0.076 0.153** 0.108 –0.029 0.085** –0.083 0.141*** 0.087 –0.082 0.035 0.148** 0.095 0.077 0.149** 0.112 –0.036 0.084** –0.083 0.141*** 0.086 –0.080 –0.014 0.149** 0.091 0.082 0.148*** 0.105 –0.030 0.084** –0.087 0.138*** 0.091 –0.081 0.026 0.153** 0.089 0.080 0.143* 0.110 –0.031 Total consumer goods Apparel, household goods–cotton Furniture, household goods Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles–non-wool or cotton Domestic prices 0.000 0.169 0.189** –0.177 0.179 0.148 0.170 –0.226 0.056 0.218 –0.097 0.258* 0.114 0.135*** 0.148* 0.152* 0.084 0.166 0.142 0.065 0.132** 0.169** 0.083** –0.066 0.141* 0.096 –0.076 –0.038 0.146* 0.112* 0.082 0.127 0.105 –0.025 Note: *,**, and *** indicate significance at the 10 percent, 5 percent, and 1 percent levels, respectively. The elasticities are computed from the estimation of the coefficient b in equation (3). 1 Source: Authors’ calculations 11. We believe the proxy we use for the prices of competing goods is not as precise in this case. Within the industrial supplies and materials category, most of the items are commodities or very standardized products, so domestic prices and import prices refer to almost the same good and are therefore highly correlated. ECONOMIC REVIEW Third Quarter 2005 29 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 8 ERPT Differentials (ERPT Depreciation Minus ERPT Appreciation) Economic models Foreign costs Statistical model Broadbased PPI/WPI– based CPIbased Domestic prices Total imports –0.228 –0.243 –0.271*** –0.249 Total industrial supplies and materials Plastic materials Organic chemicals Iron and steel mill products Finished metal shapes Crude oil Fuel oil Petroleum products, other Gas–natural Bauxite and aluminum Lumber Shingles and wallboard –0.444 0.211 –0.331 –0.090 –0.223 –0.756 –0.446 –0.312 –0.683 –0.268 –0.550* –0.518** –0.482* 0.184 –0.342* –0.110 –0.205 –0.716 –0.294 –0.320 –0.686 –0.287 –0.565* –0.520** –0.523** 0.157 –0.331 0.048 –0.224 –0.779 –0.526 –0.357 –0.680 –0.238 –0.592* –0.549** –0.487* 0.195 –0.317 –0.088 –0.242 –0.734 –0.465 –0.318 –0.688 –0.256 –0.630** –0.532** –0.362 0.184 –0.321* –0.168 –0.224 –0.445 –0.061 0.370 –0.395 –0.210 –0.083 –0.300 Total capital goods except automotive Electrical apparatus Industrial machines, other Computer accessories Computers Semiconductors Telecom equipment Medicinal equipment Photo, service machinery –0.062 0.142 0.178 –0.145 0.042 –0.144 0.150 –0.099 0.203 –0.062 0.138 0.169 –0.149 0.129 –0.122 0.183 –0.060 0.181 –0.061 0.148 0.176 –0.151 0.121 –0.100 0.151 –0.101 0.188 –0.083 0.156 0.176 –0.151 0.163 –0.117 0.146 –0.097 0.179 –0.058 0.135 0.174 –0.140 0.191 –0.095 0.082 –0.134 0.181 Total consumer goods Apparel, household goods–cotton Furniture, household goods Other household goods Toys, games, sporting goods TVs, VCRs, etc. Gems, diamonds Household appliances Footwear Pharmaceutical preparations Writing and art supplies Apparel, textiles–non-wool or cotton 0.119 –0.019 –0.150 –0.147 –0.142 –0.142 –0.234 0.224*** 0.143 0.120 –0.165 0.019 0.113 0.033 –0.147 –0.152 –0.130 –0.110 –0.215 0.213*** 0.139 0.121 –0.157 –0.043 0.119 –0.008 –0.148 –0.148 –0.140 –0.125 –0.260* 0.222*** 0.139 0.118 –0.161 –0.027 0.124 –0.044 –0.148 –0.148 –0.148 –0.152 –0.228 0.229*** 0.149 0.054 –0.147 –0.024 0.115 0.130 –0.152 –0.096 –0.140 –0.120 –0.257 0.171 0.139 0.064 –0.160 –0.034 Note: *,**, and *** indicate significance at the 10 percent, 5 percent, and 1 percent levels, respectively. The elasticities are computed from the estimation of the coefficient b4 in equation (4). Source: Authors’ calculations At the most disaggregated level we cannot reject the non-pass-through hypothesis in the majority of cases. As we point out earlier, the estimations of ERPT elasticities for industrial supplies and materials are not very robust; nevertheless, we find that the plastic materials item is consistently significant, with a degree of pass-through in the range of 19 percent to 24 percent. Within capital goods, we find statistically significant partial ERPT for several items: electrical apparatus (15 percent); industrial machines, other (15 percent); medicinal equipment (13 percent); and photo, service 30 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 3a Evolution of ERPT Coefficients for All Categories, January 1998–December 2004 Total imports Industrial supplies .08 .25 .20 .06 .15 .04 .10 .02 .05 0 0 1998 2000 2002 2004 1998 Capital goods 2000 2002 2004 Consumer goods .04 .016 .03 .012 .02 .008 .01 .004 0 0 1998 2000 2002 2004 1998 2000 2002 2004 Note: The graphs show recursive estimations of equation (3), using the PPI/WPI specification from Table 7. The thin lines show one-standarddeviation bounds. machinery (17 percent). Finally, within consumption goods, the three items that are consistently significant are furniture (14 percent); gems, diamonds (15 percent); and pharmaceutical preparations (15 percent). Table 8 shows the differential ERPT elasticities obtained from estimating equation (4). As in the rest of the literature, our study finds no evidence of asymmetric passthrough in the vast majority of cases. Thus, from our econometric exercises we conclude that the degree of pass-through is the same whether the exchange rate depreciates or appreciates, a finding that contradicts some of the preliminary ideas described earlier. In general, we cannot reject the hypothesis of zero differentials. Only three items (lumber, shingles and wallboard, and household appliances) evidence different behavior, but in all of them the ERPT coefficients are not significant, either overall or during appreciation events. Furthermore, as in the previous table, the results for the first two items, which fall in the industrial supplies and materials category, are not robust across all specifications. Finally, we estimate equation (3) using recursive least squares. This technique implies equation (3) is estimated repeatedly using a larger sample each time. We start with a sample size of t = 48 and then generate a vector of (T – 48) coefficients by adding one new observation to the sample until t = T. We report these vectors in Figures 3a–3d, which plot the path of the ERPT coefficients and one-standarddeviation bounds.12 12. The charts show the evolution of the coefficients as they come from the regression, which should be transformed to be read as elasticities. ECONOMIC REVIEW Third Quarter 2005 31 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 3b Evolution of ERPT Coefficients for Industrial Supplies, January 1998–December 2004 Plastic materials Organic chemicals Iron and steel .12 .12 .04 .08 .08 0 .04 .04 0 –.04 0 1998 –.04 2000 2002 2004 1998 2000 2002 2004 1998 2000 Crude oil Finished metal 2002 2004 Fuel oil .6 .6 .08 .4 .4 .2 .04 .2 0 1998 2000 2002 2004 0 1998 0 –.2 2000 2002 2004 1998 2000 Natural gas Petroleum products .4 2002 2004 Bauxite and aluminum .8 .08 .3 .2 .4 .04 0 –.04 0 .1 0 –.08 –.1 1998 2000 2002 2004 –.4 1998 2000 2002 Lumber 2004 1998 2000 2002 2004 Shingles and wallboard .1 .1 0 0 –.1 1998 2000 2002 2004 –.1 1998 2000 2002 2004 Note: The graphs show recursive estimations of equation (3), using the PPI/WPI specification from Table 7. The thin lines show one-standarddeviation bounds. The figures show that the degree of pass-through of total imports has a slightly downward trend during the analysis period. However, the behavior of its components is very heterogeneous. While industrial supplies and materials (Figure 3b) closely resemble the aggregate pattern (Figure 3a) over the period, the other two categories present a change in the trend in the last months of 2004, when both capital goods (Figure 3c) and consumer goods (Figure 3d) prices increase their sensitivity to exchange rate movements. The heterogeneity is more evident among the components of each category. Within industrial supplies and materials, items such as natural gas, bauxite and aluminum, and lumber have a definite upward trend. Among the components of capital goods, all but computers and medicinal equipment show slight increases in 32 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 3c Evolution of ERPT Coefficients for Capital Goods, January 1998–December 2004 Computer accessories Industrial machines, other Electrical apparatus .06 .06 .02 .04 0 .02 –.02 .04 .02 0 1998 2000 2002 2004 0 1998 2000 Computers 2002 2004 –.04 1998 2000 Semiconductors .12 2002 2004 Telecom equipment .08 .04 .04 .02 0 0 .08 .04 0 –.04 –.08 1998 2000 2002 2004 –.04 1998 –.02 2000 2002 2004 1998 2000 2002 2004 Photo, service machinery Medicinal equipment .06 .04 .04 .02 .02 0 1998 2000 2002 2004 0 1998 2000 2002 2004 Note: The graphs show recursive estimations of equation (3), using the PPI/WPI specification from Table 7. The thin lines show one-standarddeviation bounds. their ERPT coefficients during the last months, but in most of the cases the coefficients are drifting down over the whole period. In the last category, consumer goods, variables are trending down (furniture; other household goods; gems, diamonds; pharmaceutical preparations; and apparel, textiles–non-wool or cotton), up (toys, games, sporting goods and writing and art supplies), or showing no trend (apparel–cotton; TVs, VCRs, etc.; household appliances; and footwear). During the final months of 2004, however, almost all the items show a stable or an upward trend in the ERPT coefficient. Conclusion This article seeks to answer the question of why the dollar’s depreciation has not stopped the trade deficit from deepening in the past few years. Is it that the products the United States imports have not become more expensive? Or is it that even when imports are more expensive we buy them anyway? The answer seems to be yes in both cases. On the one hand, prices of capital and consumer goods have not absorbed much of the movements in the exchange rate (either depreciations or appreciations) during the past ten years and consequently have remained fairly stable. On the other hand, even though prices of industrial supplies and materials have been rising, we have continued to import them. ECONOMIC REVIEW Third Quarter 2005 33 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure 3d Evolution of ERPT Coefficients for Consumer Goods, January 1998–December 2004 Furniture Apparel–cotton Other household goods .01 .02 .04 0 .01 –.01 .02 0 –.02 1998 2000 2002 2004 0 1998 2000 2002 2004 1998 TVs, VCRs, etc. Toys, games, sporting goods 2000 2002 2004 Gems, diamonds 0 .08 .02 .06 –.01 0 –.02 .04 –.02 –.03 .02 –.04 –.04 –.06 1998 2000 2002 2004 0 1998 2000 2002 2004 1998 Footwear Household appliances 2000 2002 2004 Pharmaceutical preparations .02 .02 .06 .01 .04 0 .02 .01 0 –.01 1998 2000 2002 2004 0 –.01 1998 2000 2002 Writing and art supplies 2004 1998 2000 2002 2004 Apparel, textiles–non-wool or cotton .04 .03 .02 .02 0 .01 –.02 0 –.04 –.01 1998 2000 2002 2004 1998 2000 2002 2004 Note: The graphs show recursive estimations of equation (3), using the PPI/WPI specification from Table 7. The thin lines show one-standarddeviation bounds. A third question inevitably arises: Will this performance continue in the future? To answer this question, we turn to the analysis of the behavior of some import price indexes during the past decade. Overall, our results show that exchange rate movements are translated only slightly into changes in the domestic price of imports at a monthly frequency. The ERPT elasticity of total imports’ prices averages 18 percent although there is a considerable degree of variation across import categories. We find that capital and consumer goods consistently have low degrees of ERPT. In these categories, dollar depreciations (appreciations) appear to be matched fairly closely by reductions (increases) in the foreign price of these products. We believe this observation exposes in part the behavior of foreign exporters, suggesting that 34 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA they alter their profit margins in response to exchange rate changes. Our results also suggest that the dollar’s value does not affect either the domestic or the foreign price of the imports of industrial supplies and materials, revealing the absence of a buffering effect from foreign margins. As previous studies have found, we also find a downward trend in ERPT elasticities for the main import categories (see Taylor 2000; Yang 1997; Swamy and Thurman 1994). However, this trend is not evident at the more disaggregated levels, where a reversion toward higher ERPT may be observed during 2004. This last observation is crucial for responding to the third question. It suggests that some foreign firms have stopped absorbing exchange rate depreciations. After a long period of a falling dollar, margins have become slim, and the chances of continuing with the same strategy of price adjustment have been reduced (see Greenspan 2005). Hence, to be able to survive, some foreign exporters are likely to start passing through exchange rate depreciations to domestic import prices, and we would then see the cheaper dollar feeding into some domestic import prices. Under this scenario, our response to the third question would be “probably not.” Obviously, we are looking at just one side of the coin. While the capital account remains positive, the current account, and in turn the trade balance, will remain negative. Consequently, the dollar’s depreciation might continue, the import bundle might change, and we would still observe low ERPT into the aggregate index of domestic import prices. Appendix A Recovering ERPT Elasticities from Regression Coefficients From equation (2), the coefficient β is the elasticity of domestic import prices to the nominal exchange rate (ERPT): 1 dP d log Pt p dP E β= = = ⋅ = ηP,E . d log Et 1 dE P dE E The estimated coefficient b1 in equation (3) is b1 = d∆ log Pt d log Pt − d log Pt −1 = d∆ log Et d log Et − d logg Et−1 Pt −1 − Pt 1 1 dP − dP Pt Pt −1 Pt −1 dP Et ⋅ ⋅ . = = 1 1 Et −1 − Et dE Pt dE − dE Et Et −1 Et −1 Using the above definition for the ERPT, ∆Pt ∆P E P b1 = t −1 ⋅ ηP , E = t ⋅ t −1 ⋅ ηP , E ≅ ( ηP , E )2 . ∆Et ∆Et Pt −1 Et −1 ECONOMIC REVIEW Third Quarter 2005 35 F E D E R A L R E S E R V E B A N K O F AT L A N TA Appendix B Deriving a Cost Proxy from the IFS Exchange Rate Series The IFS provides real effective exchange rate (REER) based on unit labor cost. The index is defined as the nominal exchange rate times a ratio of unit labor costs: (B1) reu = neu ⋅ ω* , ω hw* = where reu is the REER adjusted by labor costs, neu is the nominal exchange rate, and ω * and ω are the foreign and domestic normalized unit labor costs, respectively. These costs are defined as the ratio of hourly compensation in manufacturing to measured labor productivity in that sector: (B2) ω = hw ; ω* = ω, for the country’s entire manufacturing sector. The IFS reports this index for several countries, based on data availability. Inserting equations (B2) into (B1) and rearranging the terms, we obtain hw * * reu ⋅ hw ⋅ neu * . If we assume that the ratio of productivities among the United States and its major trading partners is not significantly altered during the period under study (normalized to 1), then * =1. Thus, it is straightforward to obtain an expression to estimate the proxy of the exporter’s foreign costs: , where hw is the hourly wage and l is the measure of productivity in each sector. Adding up all the sectors, it is possible to obtain an index, hw* = reu ⋅ hw. neu REFERENCES Baldwin, Richard, and Paul R. Krugman. 1989. Persistent trade effects of large exchange rate shocks. Quarterly Journal of Economics 104, no. 4:635–54. Bernard, Andrew, and Bradford Jensen. 2004. Entry, expansion, and intensity in the U.S. export boom, 1987–1992. Review of International Economics 12, no. 4:662–75. Burstein, Ariel T., João C. Neves, and Sergio Rebelo. 2003. Distribution costs and real exchange rate dynamics during exchange-rate-based stabilizations. Journal of Monetary Economics 50, no. 6:1189–1214. Campa, Jose M., and Linda Goldberg. 2002. Exchange rate pass-through into import prices: A macro or micro phenomenon? NBER Working Paper Nos. 8934, May. Dornbusch, Rudiger. 1987. Exchange rates and prices. American Economic Review 77, no. 1:93–106. Feenstra, Robert C. 1989. Symmetric pass-through of tariffs and exchange rates under imperfect competition: An empirical test. Journal of International Economics 27, no. 1:25–45. Goldberg, Pinelopi K., and Michael M. Knetter. 1997. Goods prices and exchange rates: What have we 36 ECONOMIC REVIEW Third Quarter 2005 learned? Journal of Economic Literature 35, no. 3:1243–72. Greenspan, Alan. 2005. Current account. A speech at the Advancing Enterprise 2005 Conference, London, February 4. Marazzi, Mario, Nathan Sheets, and Robert J. Vigfusson. 2005. Exchange rate pass-through to U.S. import prices: Some new evidence. Board of Governors of the Federal Reserve System, photocopy. Olivei, Giovanni P. 2002. Exchange rates and the prices of manufacturing products imported into the United States. New England Economic Review (First Quarter): 3–18. Swamy, P.A.V.B., and Stephan S. Thurman. 1994. Exchange rate episodes and the pass-through of exchange rates to import prices. Journal of Policy Modeling 16, no. 6:609–23. Taylor, John B. 2000. Low inflation, pass-through, and the pricing power of firms. European Economic Review 44, no. 7:1389–1408. Yang, Jiawen. 1997. Exchange rate pass-through in U.S. manufacturing industries. Review of Economics and Statistics 79, no. 1:95–104. F E D E R A L R E S E R V E B A N K O F AT L A N TA It’s Who You Are and What You Do: Explaining the IT Industry Wage Premium JASON DEBACKER, JULIE HOTCHKISS, MELINDA PITTS, AND JOHN ROBERTSON DeBacker is a Ph.D. candidate at the University of Texas at Austin. Hotchkiss is a research economist and policy adviser, Pitts is a research economist and associate policy adviser, and Robertson is an assistant vice president, all in the regional group of the Atlanta Fed’s research department. he investment in and use of information technology (IT) was undoubtedly an important contributor to the rapid growth of the U.S. economy during the 1990s. By one estimate the IT-producing sector was responsible for 1.4 percentage points of the nation’s average annual real gross domestic product (GDP) growth of 4.6 percent between 1996 and 2000 (Economics and Statistics Administration 2003). But in 2001 the situation changed dramatically as business spending on IT equipment and services declined, and in 2002 IT-producing industries contributed only an estimated 0.1 percentage points to the 2 percent real GDP growth. A recent paper by Hotchkiss, Pitts, and Robertson (2005) documents the wage outcomes for workers during and after the IT boom of the 1990s using a unique set of employer-employee matched earnings data for workers in Georgia. One of the paper’s findings is that, after controlling for individual characteristics, workers in IT-producing industries have average earnings that are much higher than those in other industries. Workers in IT service industries, in particular, accrue a relatively large wage premium. Hotchkiss, Pitts, and Robertson speculate that these different wage outcomes may be related to the types of occupations IT workers hold across industries. Unfortunately, the data used in their paper do not contain information on a worker’s occupation. This article’s main objective is to present evidence on the extent to which variation in average wages between IT-producing and non-IT industries can be accounted for by differences in wages paid to IT-related occupations.1 If average industry wage differentials in IT-producing industries are substantially lower after controlling for IT occupation, this finding will reinforce the notion that occupation wage differentials are an important source of the observed wage premium accruing to workers in IT-producing industries. The article first describes the data used in the analysis and then discusses the various estimates of the average industry wage differentials. The sample average wage differences across industries are compared with the differences obtained after T ECONOMIC REVIEW Third Quarter 2005 37 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 1 IT-Related Occupations in the Current Population Survey Billing, posting, and calculating machine operators Electrical power installers and repairers Broadcast equipment operators Electronic repairers, communications and industrial equipment Communications equipment operators, n.e.c. Office machine operators, n.e.c. Computer operators Office machine repairers Computer programmers Computer systems analysts and scientists Operations and systems researchers and analysts Data-entry keyers Peripheral equipment operators Data processing and equipment repairers Supervisors, computer equipment operators Electrical and electronic engineers Telephone installers and repairers Electrical and electronic equipment assemblers Telephone line installers and repairers Electrical and electronic technicians controlling for category of occupation and those obtained after controlling for occupation as well as individual and geographical characteristics. The article closes with some conclusions about the IT wage premium. About the Data The data used in this study come from the Current Population Survey’s Earner Study conducted by the Census Bureau and the Bureau of Labor Statistics. To incorporate variation in the IT industries associated with the employment boom of the late 1990s and the subsequent bust, data from surveys for 1996 to 2002 are used.2 The sample includes U.S. workers aged eighteen to sixty-four who are not selfemployed, who work in private industries not based on natural resources, who have positive wages that do not exceed $150 per hour in 2002 dollars, and for whom no data are missing. The resulting sample is a set of seven annual cross sections with a total of 845,045 observations. All observations within a year represent unique individuals. An individual’s industry and occupation cohort is defined according to the individual’s primary job. The categories used to define cohorts reflect the type of occupation, IT-related or non-IT-related, and nine industry or sector groups. The twenty job descriptions that identify a worker as having an IT-related occupation (listed in Table 1) are based on those used by the Economics and Statistics Administration (1999). The definition of the IT-related occupation category includes a broader array of jobs than what might be considered “core” IT jobs such as computer scientists, engineers, programmers, and system analysts. Specifically, the category also includes jobs deemed important to maintaining the infrastructure of the IT-producing industries—for example, data-entry keyers, telephone installers, and equipment repairers. The industry groupings are taken from the 1990 Census of Population Industrial Classification System (Census Bureau), with the IT-producing sector defined as in Economics and Statistics Administration (1999). To focus on IT-producing industries in more detail and to make them comparable to the IT industry classification used in Hotchkiss, Pitts, and Robertson (2005), we divide the IT-producing sector into three components: (1) the manufacturing of IT equipment or components, (2) communication services, and (3) software and computer services. The non-IT industries are con- 38 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Figure Wages by Industry and Occupation IT manufacturing IT communication services 2002 $/hour 26 26 22 22 18 18 14 14 14 10 10 IT workers 22 Non-IT workers 18 1996 1998 2000 2002 10 1996 Construction 2002 $/hour 1998 2000 2002 1996 Non-IT manufacturing 26 22 22 18 18 18 14 14 14 10 1998 2000 2002 1998 2000 2002 1996 Finance, insurance, and real estate 26 26 22 22 22 18 18 18 14 14 14 10 1996 1998 2000 2002 1998 2000 2002 Other non-IT services 26 10 2002 10 1996 Wholesale and retail trade 2000 26 22 1996 1998 Transportation and utilities 26 10 2002 $/hour Software and computer services 26 10 1996 1998 2000 2002 1996 1998 2000 2002 Source: CPS Earner Study struction; non-IT manufacturing; transportation and utilities; wholesale and retail trade; finance, insurance, and real estate; and miscellaneous non-IT services. The combination of industry and occupation makes up eighteen industry/occupation cohorts. The figure presents the time path of average real wages across the nine industries over the 1996–2002 period for IT and non-IT occupations. The charts show that the average wage of IT occupations is greater than for non-IT occupations irrespective of industry.3 The average wage of IT occupations across all industries is $20.62, and the average for non-IT occupations is $15.02. 1. Comparing relative wage outcomes for transitioning workers after controlling for occupation is left to future research. 2. Because of changes in occupational and industry classifications associated with the shift from Standard Industrial Classification to the North American Industrial Classification System definitions, data for years after 2002 are not directly comparable to earlier years and so are excluded from the analysis. 3. The charts also show that the occupation wages vary by industry. For instance, the average wage of IT workers is $22.08 in the IT-producing sector as a whole but only $19.26 in non-IT industries. At the same time, occupational wage differentials tend to vary across industry. For instance, a large wage gap separates IT and non-IT occupations in the non-IT manufacturing industry, but only a relatively small wage difference appears in IT manufacturing. Separate regression analysis shows that controlling for these sources of variation by interacting occupation and industry does not change the basic findings regarding the industry wage differentials. ECONOMIC REVIEW Third Quarter 2005 39 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 2 The Share of Workers in IT-Related Occupations by Industry IT manufacturing Communication services Software and computer services Construction Non-IT manufacturing Transportation and utilities Wholesale and retail trade Finance, insurance, and real estate Miscellaneous non-IT services Percent of IT workers Total number of workers 31.8 33.0 43.5 1.2 3.8 5.3 1.2 6.6 2.8 21,270 16,360 21,775 55,039 146,172 46,787 196,711 66,807 274,124 Source: CPS Earner Study Table 2 presents the average share of workers in IT-related occupations by industry and the total number of workers. In the sample, 32 percent of the workers in IT manufacturing, 33 percent of those in communication services, and 44 percent of those in software and computer services are in core IT occupations. For the nonIT industries the concentration of IT workers is much lower, ranging from 7 percent of workers in finance, insurance, and real estate to 1 percent of workers in construction and in wholesale and retail trade. Analysis of the Data: Industry Wage Differentials The primary focus of the analysis is on the average wage differentials across industries and whether these differences can be accounted for by occupation over and above other worker characteristics. The sample average industry wage differentials presented in the first column of Table 3 are obtained from a regression of the logarithm of the real hourly wage on a set of dummy variables that identify the worker’s industry of employment (using non-IT manufacturing as the reference industry).4 As in Hotchkiss, Pitts, and Robertson (2005), the highest average earnings are in the IT-producing sector. The average wage in the software and computer services industry is 27.4 percent higher than that for non-IT manufacturing while in communication services and IT manufacturing the wage gap is 21.4 and 17.3 percent, respectively. Average wages in finance, insurance, and real estate; transportation and utilities; and construction are 7.0, 4.5, and 0.5 percent higher than in non-IT manufacturing, respectively. In contrast, workers in the wholesale and retail trade sector earn an average 32.8 percent less, and average wages in miscellaneous non-IT service industries are 11.9 percent less than in non-IT manufacturing. Controlling for occupation. The fact that workers in the IT sector have high average wages is not surprising. The figure shows that workers in IT-related occupations earn more on average than non-IT workers while Table 2 shows that IT industries have a large concentration of workers in IT-related occupations. To estimate the industry wage differentials after controlling for IT occupations, the logarithm of real hourly wage is regressed on a dummy variable equal to 1 for an IT occupation and 0 otherwise in addition to the set of dummy variables for industry of employment. The estimation results are presented in the second column of Table 3. 40 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 3 Average Occupation and Industry Wage Differentials Unconditional IT occupation Controlling for occupation Controlling for individual characteristics Controlling for occupation and individual characteristics 0.1884 (0.0031) 0.0925 (0.0026) IT manufacturing 0.1732 (0.0043) 0.1204 (0.0043) 0.0656 (0.0036) 0.0404 (0.0036) Communication services 0.2143 (0.0048) 0.1593 (0.0049) 0.1041 (0.0040) 0.0777 (0.0041) Software and computer services 0.2741 (0.0042) 0.1993 (0.0044) 0.1051 (0.0036) 0.0697 (0.0037) Construction 0.0047 (0.0029) 0.0096 (0.0029) 0.0458 (0.0024) 0.0484 (0.0024) Transportation and utilities 0.0450 (0.0031) 0.0421 (0.0031) 0.0007 (0.0026) –0.0005 (0.0026) Wholesale and retail trade –0.3279 (0.0020) –0.3231 (0.0020) –0.1682 (0.0017) –0.1663 (0.0017) Finance, insurance, and real estate 0.0702 (0.0027) 0.0649 (0.0027) 0.0276 (0.0023) 0.0251 (0.0023) Miscellaneous non-IT services –0.1192 (0.0019) –0.1174 (0.0019) –0.0848 (0.0017) –0.0842 (0.0017) Note: These percent average wage differentials are relative to workers in non-IT manufacturing. The numbers in parentheses are standard errors. The estimated IT occupation differential is 18.8 percent. That is, given the industry of employment, someone in an IT-related occupation is expected to earn 18.8 percent more than someone in a non-IT occupation. Because the share of IT workers in the IT sector is much greater than in non-IT industries, including the occupation identifier lowers the average wage differential in IT industries much more than in non-IT industries. Nonetheless, average IT-industry wage premiums remain quite large (12.0 percent in IT manufacturing, 15.9 percent in communication services, and 19.9 percent in software and computer services), suggesting that factors other than simply identifying the worker as having an IT-related occupation are important. Controlling for individual characteristics. Individual worker characteristics not accounted for by IT occupation may explain some of the remaining wage variation across industries. For instance, the relatively high average pay of workers in the software and computer services industry might be attributable to the fact that all workers in this industry are disproportionately more highly educated and that a general wage premium exists for more education.5 4. All the regressions in this section also include a set of dummy variables that identify the year to control for covariation over time. 5. See, for example, Hellerstein, Neumark, and Troske (1999) for some recent evidence on the relative importance of individual characteristics for wage differences across industries. ECONOMIC REVIEW Third Quarter 2005 41 F E D E R A L R E S E R V E B A N K O F AT L A N TA Table 4 Individual Characteristic Estimation Results Controlling for industry Controlling for industry and occupation Age 0.0501 (0.0003) 0.0500 (0.0003) Age squared –0.0005 (0.0000) –0.0005 (0.0000) Less than high school education –0.1951 (0.0019) –0.1941 (0.0019) Some college, no degree 0.1205 (0.0013) 0.1191 (0.0013) College degree or higher 0.4466 (0.0015) 0.4446 (0.0015) Female –0.1850 (0.0012) –0.1831 (0.0012) Black –0.1773 (0.0019) –0.1770 (0.0019) Hispanic –0.1803 (0.0020) –0.1798 (0.0020) Other race –0.1114 (0.0025) –0.1143 (0.0025) Part-time –0.1810 (0.0015) –0.1802 (0.0015) Union 0.1332 (0.0019) 0.1333 (0.0019) Midwest –0.0088 (0.0016) –0.0087 (0.0016) South –0.0200 (0.0016) –0.0200 (0.0016) West 0.0170 (0.0016) 0.0174 (0.0016) Nonmetro area –0.2006 (0.0016) 0.1995 (0.0016) Metro size 100,000–249,999 –0.1457 (0.0024) –0.1450 (0.0024) Metro size 250,000–499,999 –0.1237 (0.0022) –0.1232 (0.0022) Metro size 500,000–999,999 –0.0980 (0.0020) –0.0974 (0.0020) Metro size 1,000,000–2,499,999 –0.0765 (0.0017) –0.0763 (0.0017) Metro size 2,500,000–4,999,999 –0.0216 (0.0024) –0.0220 (0.0024) 0.3590 0.3600 Adjusted R-squared Note: The first column of estimates refers to the model without the occupation control. The second column of estimates refers to the model with the occupation control included. The numbers in parentheses are standard errors. 42 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA To control for the effect of individual characteristics on wages, the logarithm of the real hourly wage is regressed on the industry dummy variables and a set of individual characteristics: age, educational attainment, gender, race, geographical location, union status, and part-time work status. The estimated wage differentials are reported in the third column of Table 3. The estimated coefficients on the individual characteristics are quite consistent with standard human capital theory. For example, the coefficients on age and age squared show that earnings increase with experience but at a decreasing rate. At the Even after human capital differences and sample mean age of 37.99, holding other differences that arise across occupations factors constant, an extra year of age adds are controlled for, workers in IT-producing 1.2 percent to expected wages. More important to earnings than age is educaindustries still enjoy a wage premium over tional attainment. Workers with at least a workers in other sectors. college degree earn 44.7 percent more than those with a high school diploma, other factors held constant. Female workers earn 18.5 percent less than male workers, and black and Hispanic workers each earn about 18 percent less than white workers. Union workers earn 13.3 percent more than nonunion workers. Part-time workers earn 18.1 percent less than full-time workers, and workers in the South and the Midwest earn 2 and 0.9 percent less than those in the Northeast, respectively, while those in the West earn 1.7 percent more than those in the Northeast. A comparison of the third column of Table 3 with the first and second columns shows that controlling for individual characteristics reduces the estimated average industry wage differentials across all industries. Further, for IT-producing industries, the reduction is by more than would be seen by simply controlling for occupation. For IT manufacturing the estimated average wage premium is 6.6 percent. For software and computer services the premium is 10.5 percent, and for communication services, 10.4 percent. Controlling for occupation and individual characteristics. The remaining question is whether controlling for IT occupation results in a further reduction in the average industry wage differentials after controlling for individual characteristics. To control for individual characteristics as well as occupation, the logarithm of the real hourly wage is regressed on the IT occupation dummy variable, the industry dummy variables, and the set of individual characteristics. The estimated occupation and wage differentials are reported in the fourth column of Table 3. The estimated coefficients on the individual characteristics are similar to those obtained when occupation is excluded from the regression and are reported in Table 4. Comparing the fourth and second columns of Table 3 shows that controlling for individual characteristics reduces the size of the IT-occupation wage premium from 18.8 percent to 9.3 percent. Comparing the fourth and third columns in Table 3 shows that including the occupation identifier does matter for the average industry wage differentials in the IT-producing sector but does not matter as much as do the individual characteristics. Specifically, including occupation reduces the average industry wage differentials in the IT-producing industries by between 2.5 and 3 percentage points. Across industries, the average industry wage differentials are all less than 10 percent except for the wholesale and retail trade sector; however, the largest premiums still accrue to workers in IT-producing industries.6 ECONOMIC REVIEW Third Quarter 2005 43 F E D E R A L R E S E R V E B A N K O F AT L A N TA Conclusions Working in an IT industry is associated with higher-than-average wages. Both IT and non-IT workers in the IT-producing sector (as broadly defined in this article) are paid more on average than their counterparts in the various non-IT sectors. Part of the reason for the high wages in IT-producing industries appears to be that the average wage of IT occupations is greater than for non-IT occupations, and IT-producing industries have a disproportionately large share of their workforce in IT-related occupations. In other words, the IT industry wage differentials are partly attributable to occupation wage differences. Controlling for individual worker characteristics such as gender, race, education, part-time status, and location is also very important and substantially lowers the average industry wage premium across all industries. However, accounting for individual characteristics reduces, but does not eliminate, the IT occupation effect on IT industry wages and further reduces the wage premium in IT-producing industries. These findings are broadly consistent with those in Hotchkiss, Pitts, and Robertson (2005) and suggest that workers in IT-producing industries generally have high levels of human capital. But even after human capital differences (through inclusion of individual characteristics) and differences that arise across occupations are controlled for, workers in IT-producing industries still enjoy a wage premium over workers in other sectors. 6. The IT occupation grouping used here is too coarse to capture the effect on wages of differences in the distribution of IT occupations across industries. For example, an examination of the distribution of the twenty IT-related occupations in the communication services industry shows that the most populous occupation is telephone installer and repairer (36 percent of IT workers versus 9.8 percent for the IT sector overall). Interestingly, these workers are much older than average for IT workers (40.8 years versus 37.8 years), but they have much less formal education (only 10.7 percent have a college degree compared with 39.9 percent for the IT sector as a whole) and are more unionized (51.7 percent versus 9.6 percent) than other IT workers in the IT-producing sector. This observation suggests that blue-collar IT service workers, perhaps because of the need for longterm, on-the-job training such as apprenticeships, receive a wage premium that is not adequately accounted for by the broad IT-occupation categorization. An extension of the wage model used here would be to control for blue-collar versus white-collar IT occupations. 44 ECONOMIC REVIEW Third Quarter 2005 F E D E R A L R E S E R V E B A N K O F AT L A N TA REFERENCES Economics and Statistics Administration. 1999. The Emerging Digital Economy II. Washington, D.C.: U.S. Department of Commerce, June. ———. 2003. Digital Economy 2003. Washington, D.C.: U.S. Department of Commerce, December. Hellerstein, Judith K., David Neumark, and Kenneth R. Troske. 1999. Wages, productivity, and worker characteristics: Evidence from plant-level production functions and wage equations. Journal of Labor Economics 17, no. 3:409–46. Hotchkiss, Julie L., M. Melinda Pitts, and John C. Robertson. 2005. Earnings on the information technology rollercoaster: Insight from matched employeremployee data. Federal Reserve Bank of Atlanta Working Paper No. 2005-11, June. ECONOMIC REVIEW Third Quarter 2005 45