What Drives Gold Prices? - Federal Reserve Bank of Chicago (2024)

A half century after gold ceased to play a significant formal role in theinternational monetary system, it still captures a great deal of attention inthe financial press and the popular imagination. Yet there has been verylittle scrutiny of the primary factors determining the price of gold since itsdollar price was first allowed to vary freely in 1971.1 In thisarticle, we attempt to fill in that gap by highlighting three considerationsthat are commonly cited as drivers of gold prices: inflationary expectations,real interest rates, and pessimism about future macroeconomic conditions.

Our empirical results in this Chicago Fed Letter are organized aroundthree claims—namely, that gold is a hedge against inflation, gold is sensitiveto expected long-term real interest rates, and gold is regarded as protectiveagainst “bad economic times.”

Gold is a hedge against inflation. A rise in inflation orinflationary expectations increases investors’ interest in purchasing goldand, therefore, drives up its price; in contrast, disinflation or a drop ininflationary expectations does the opposite. We will measure the “inflationhedge” motive for holding gold with PTR—which is the mnemonic for thesurvey-based ten-year inflation expectation that is provided by the Board ofGovernors of the Federal Reserve System; PTR has in recent years coincidedwith the ten-year inflation projection of theSurvey of Professional Forecasters (SPF) conducted by the FederalReserve Bank of Philadelphia.2 The notion that gold can beidentified with an inflation protection motive is of course connected with thefact that, in contrast to fiat money, gold is in nearly fixed supply. But thisproperty of gold is shared by many other commodities. The special statusaccorded gold may be a relic of the gold standard era, or it may even reflecta belief on the part of a subset of investors that there is a positiveprobability that the world will at some point return to a gold standard.Figure 1 shows how the real price of gold and the long-term inflationexpectation have evolved over time. The measure of the real gold price is theLondon PM fixing price for gold (from the London Bullion Market Association)in U.S. dollars per ounce deflated by the U.S. Consumer Price Index, or CPI(from the U.S. Bureau of Labor Statistics), plotted on a log scale; and themeasure of expected inflation over the next ten years is PTR. From 1971 toaround 2000, the real gold price and the long-term inflation expectation tendto move together. A sharp uptick in inflation expectations during the period1971–80 coincides with a dramatic run-up in gold prices. Gold prices felldramatically during the Volcker disinflation of 1980–83.3 Over theperiod 1983–2000, the steady downward march of expected long-term inflationfollowing the Volcker disinflation period coincides with the decrease in thereal gold price. Since 2000, however, the long-term inflation expectation hasdeviated relatively little from 2%, whereas the real gold price has increasedmore than fivefold. The role of expected inflation in this later period seemsto have given way to that of the real interest rate—our second key driver ofthe gold price—which we discuss next.

1. Real price of gold and ten-year inflation expectation, 1971:Q1–2021:Q1

Gold is sensitive to expected long-term real interest rates. Giventhat gold is a long-duration durable asset with a relatively stable dividendyield, its price is expected to have a strong inverse relationship with thelong-term real interest rate. A rise in expected real rates, all else beingequal, should drive down the price of gold.4 Figure 2 shows thereal gold price (the U.S. dollar price per ounce deflated by the CPI, onceagain on a log scale), along with the real ten-year U.S. Treasury yield (thenominal yield on ten-year Treasury securities minus PTR). The predictednegative co-movement of the real interest rate and the real gold price doesnot show up in these data before 2001.5 By contrast, between 2001and 2012, the long-term real interest rate fell some 400 basis points,accompanied by an over fivefold rise in the real gold price.

2. Real price of gold and real ten-year U.S. Treasury yield, 1971:Q1–2021:Q1

Gold is regarded as protective against “bad economic times.”  We testfor this factor’s importance by using theSurveys of Consumers conducted by the University of Michigan(Michigan survey); one of the key survey questions is the following: “Lookingahead, which would you say is more likely—that in the country as a whole we’llhave continuous good times during the next 5 years or so, or that we will haveperiods of widespread unemployment or depression, or what?”6 We useas our measure the fraction of pessimistic responses to this question, andrefer to it as “pessimistic expectations” in our analysis. Figure 3 shows thelog real gold price along with the fraction of respondents to the Michigansurvey who expect the next five years to be characterized by mostly bad times;there is considerable positive correlation between these two variables overour sample period.

3. Real price of gold and pessimistic expectations for the U.S. macroeconomy,1971:Q1–2021:Q1

Multiple regressions

Comparing figures 1–3 reveals that the key factors driving gold pricevariation often move together. For example, the rather steady rise inpessimistic expectations (figure 3) between 2001 and 2012 matches apersistently falling real interest rate over the same period (figure 2). Todisentangle the roles of the various factors over time, we perform multipleregressions.7 Our regressions provide a simple econometricevaluation of the contribution of our three key factors to the time-seriesvariation in the real gold price over the period 1971–2021. In addition, weshow that one additional factor proxied by real world or U.S. gross domesticproduct (GDP) plays an important role in accounting for the long-run trend ingold prices.

We begin with regressions that explain the association between the averageannual log level of real gold prices and four variables, also at the averageannual level: 1) the real U.S. dollar value of world GDP provided by the WorldBank, 2) the expected ten-year real interest rate computed as the nominalten-year U.S. Treasury yield minus the Federal Reserve Board’s PTR, 3) PTRitself, and 4) the fraction of the Michigan survey’s participants expectinglargely bad economic times over the next five years (i.e., the pessimisticexpectations variable). These regressions highlight the sources of longer-termvariation in the level of real gold prices over the past half century (seefigure 4). Although we find this exercise to be the most revealing about thebasic historical movements of gold prices, the sample is not large and, moreimportantly, the degree of persistence in the error term is substantial, asindicated by the relatively low Durbin–Watson statistic of 0.98.8The second regression exercise (whose results are reported in figure 5) usesessentially the same variables; but instead of looking at levels, it looks atthe relationship between the log change in the real gold price andnews about the explanatory variables using quarterly data. Finally,we conduct a limited investigation using daily data (whose regression resultsare reported in figure 6). The precise variables discussed here are notavailable at the daily frequency. However, we can investigate the roles ofexpected real rates and expected inflation using daily data on TreasuryInflation-Protected Securities (TIPS) and break-even inflation rates9 relative to nominal Treasuryyields. In these three exercises, as in all regressions on nonexperimentaldata, it is important to repeat the usual caveat that the statistical analysisreveals correlations in the data, but does not in itself establish causality.The extent to which such regressions go beyond mere association depends on the“reasonableness” of the coefficients (see note 7) and, in short, the abilityto “tell the story” that goes with the regressions.

Figure 4 shows the annual regression results. The real world GDP measure,which comes in highly significantly, reflects the fact that the demand for theservices of gold and the demand for other goods increase together,approximately one-for-one in percentage terms. The estimated coefficient onthe ten-year Treasury yield minus PTR indicates that a percentage point risein the long-term real interest rate lowers the real gold price by 13.1%. PTRhas an additional effect over and above its presence as a component of thereal rate—and indeed this is far stronger quantitatively. Given the long-termreal interest rate, an extra percentage point of ten-year expected inflationraises the real gold price by a hefty 37%—well in line with the long-held“inflation hedge” view. Finally, evaluated at the mean of 0.46, a one standarddeviation increase in the fraction of pessimistic survey respondents (8.1percentage points) raises the gold price by 9.7%.

For figure 5, we shift our focus to quarterly data. Here the conceptualexperiment is to ask how news about the explanatory variables isreflected in contemporaneous changes in the log real gold price. In additionto the markedly reduced concern about serially correlated errors, this hassomewhat more of a causal feel than the levels regression in figure 4,although the coherent story told by the levels regression gives it moreeconomic credibility than it would have on its purely econometric meritsalone. For the exercise whose results are reported in figure 5, we replace theworld output series with real U.S. GDP, in logs, given that our world GDPseries is only available annually. The news variables are constructed byrunning four predictive regressions—collectively called a vectorautoregression (VAR)—on the explanatory variables; theinnovations from this VAR constitute the news (or surprise) componentof the key explanatory variables.10 A 1% innovation in log realU.S. GDP is associated with a rise in the real gold price of 0.4%,substantially lower than the 1.1% value in the first row in figure 4, althoughin figure 5 the coefficient is very imprecisely estimated (indeed notstatistically significant). A 1 percentage point innovation in the expectedten-year real interest rate (the nominal yield on ten-year Treasury securitiesminus PTR) is associated with a 3.4% reduction in real gold prices. Instriking contrast with the result in figure 4, after accounting for the realinterest rate, innovations in PTR play no significant role in the gold price.The coefficient on innovations in the pessimistic expectations variableappears small, but this is deceptive because of the large units in which thepessimistic expectations variable is measured, as well as the large variationin this variable over time. A 10 percentage point innovation in the fractionof survey participants who expect the next five years to constitute mostly badtimes raises the real gold price by 5%. Because the pessimistic expectationsvariable repeatedly reaches lows of about 30% and highs of 60%, over theentire sample it drives substantial fluctuations in the real gold price.

Finally, we do a limited exercise using daily data and report the results infigure 6. Because the CPI is published only monthly, the dependent variable isthe daily change in the nominal gold price. This is less problematicthan it may at first appear because if we could observe daily changes in theoverall price index, they would be at least two orders of magnitude less thanthe corresponding changes in the highly volatile nominal gold price. Of theindependent variables we study in this article, only measures of the realyield on long-term Treasury securities and expected long-term inflation—inthis case taken from the TIPS market—are available at a daily frequency.However, we regard this as useful for two reasons. First, the regression isrun on the daily differences in the log nominal gold price; innovations inreal GDP or pessimistic views on the next five years are likely to beessentially constant at this frequency. Second, the roles of expected realinterest rates and inflation have been our most central theme (as evidenced bythe coefficients in figures 4 and 5), and we have the data to obtain at leastsome evidence on these at the daily frequency. Since the variables are indifferences, which are quite noisy, the R-squared, which measures the fractionof the variance of the dependent variable that is explained by the regression,is only 0.012. Yet, there are valuable lessons in this exercise. First, thenegative effect of the real interest rate on the gold price—the propositionthat comes most directly from economic theory—is once again confirmed. Hence,it has been shown to hold in annual levels, quarterly innovations, and dailydifferences. Second, the observation that the inflation effect isquantitatively much larger than the real interest rate effect holds here, aswas the case in the levels regression of figure 4, though contrary to theinnovations regression of figure 5.

Conclusion

We have investigated several hypotheses about the determinants of goldprices—in annual levels data, quarterly data in innovations form, and dailydata in differences. The negative effect of real interest rates on gold pricespredicted by theory holds in all three contexts. Two of the threespecifications (the quarterly innovations specification being the exception)support the notion that gold is an inflation hedge and that this effect isquantitatively larger than the real interest rate effect. The twospecifications that can be used to evaluate the proposition that gold pricesalso reflect protection against bad economic times are highly supportive ofit. In the early part of the sample, variation in inflation or inflationaryexpectations was the single most important consideration for the real price ofgold. From 2001 on, however, long-term real interest rates and pessimism aboutfuture economic activity appear as the dominant factors. While disinflationsince 2001 might have been expected to result in low gold prices, any effectof low inflation was more than compensated for by unprecedentedly lowlong-term real interest rates and by pessimism about future economic activity.

Notes

¹ The Bretton Woods system—which pegged the U.S. dollar price ofgold and, for the most part, fixed ratios between gold and the other maincurrencies—collapsed in stages because of inherent contradictions in thedesign of the system. In 1971, the U.S. Gold Window was closed and the fixedprice of gold vis-à-vis the dollar ended. We thus begin our sample in 1971.For a full explanation, see Michael Bordo, 2017, “The operation and demise ofthe Bretton Woods system: 1958 to 1971,” VoxEU.org, April 23,available online.

² PTR is from the Federal Reserve Board’s FRB/US model’s database;see note 4 of John M. Roberts, 2018, “An estimate of the long-term neutralrate of interest,” FEDS Notes, Board of Governors of the FederalReserve System, September 5.Crossref

³ Further details on the U.S. disinflation period of the early1980s associated with former Federal Reserve Chair Paul Volcker are in MichaelD. Bordo and Athanasios Orphanides, 2013, “Introduction,” inThe Great Inflation: The Rebirth of Modern Central Banking, MichaelD. Bordo and Athanasios Orphanides (eds.), Chicago: University of ChicagoPress, pp. 1–22,available online.

⁴ This idea manifests itself in at least two ways. First, for theowner of a gold mine to be indifferent between keeping gold in the ground onthe one hand and mining it and investing the proceeds in financial assets onthe other, the price must be expected to rise at the rate of interest. Givenan appropriate terminal condition, the higher the expected real interest rate,the lower the initial price would have to be. A second approach would be toimagine that gold provides some service flow (e.g., its value as jewelry). Thepresent value of that “dividend stream” depends inversely on the real interestrate.

⁵ This is in contradiction with Barsky and Summers (1988), whofound a strong negative correlation between the real gold price and theirmeasure of the real interest rate, particularly over the period 1973–82;rather than using survey-based inflation expectations, they used a statisticalmodel of inflation that was more sensitive to current inflation and thusprovided a quite different series for expected long-term inflation. See RobertB. Barsky and Lawrence H. Summers, 1988, “Gibson’s paradox and the goldstandard,” Journal of Political Economy, Vol. 96, No. 3, pp. 528–550.Crossref

⁶ The full Michigan survey questionnaire isavailable online.

⁷ Multiple regressions are statistical exercises estimating theeffects of several independent variables on a dependent variable. Eachregression coefficient represents the mean change in the dependent variablefor a one-unit change in the independent variable while holding constant theother independent variables.

⁸ The Durbin–Watson statistic—which measures the degree ofpersistence or serial correlation in the residuals (differences between theobserved values and the values predicted by the regression model)—takes on avalue close to 2 in the ideal case where the residuals are seriallyuncorrelated. A value close to zero indicates that the errors are sopersistent that the regression is “spurious” (uninterpretable and effectivelymeaningless). The Durbin–Watson statistic of 0.98 in the current regressionexceeds the level at which the regression would be regarded as spurious, butraises some questions about how well specified the regression is—an issuelargely addressed by the innovations formulation in figure 5. In addition, thestandard errors of the coefficients in figure 4 have been corrected for serialcorrelation as indicated in that figure.

⁹ The TIPS yield, as noted on the Federal Reserve Board’swebsite, is a real rate. The break-even inflation rate is the one that would inprinciple make a risk-neutral investor indifferent between holding a nominalTreasury security and a TIPS of the same duration. It is often regarded as ameasure of inflationary expectations at the relevant horizon.

¹⁰ A VAR is a statistical model used to capture the dynamicrelationship between two or more time-series variables; in a VAR, eachvariable is a linear function of past lags of itself and past lags of theother variable(s). In a VAR context, an innovation is the difference betweenthe observed value of a variable at a particular point in time and the optimalforecast of that value based on information available before that point intime.

Opinions expressed in this article are those of the author(s) and do notnecessarily reflect the views of the Federal Reserve Bank of Chicago or theFederal Reserve System.