Gibson’s Paradox and the Gold Standard Robert B. Barsky, Lawrence H. Summers. The Journal of Political Economy, Volume 96, Issue 3 (Jun., 1988), 528-550. Stable URL: hup:/Minks jstor-org/sici?sici=0022-3808%28 198806%2996%3A 3%3C528%3AGPATGS%3E2.0.CO%3B2-I ‘Your use of the ISTOR archive indicates your acceptance of ISTOR’s Terms and Cooditions of Use, available at bbtp:/www.jstor.org/about/tenms.htnl. ISTOR’s Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the ISTOR archive only for your personal, non-commercial use. Each copy of any part ofa JSTOR transmi printed page of such transmission, jon must contain the same copyright notice that appears on the sereen or The Journal of Political Economy is published by The University of Chicago Press. Please contact the publisher for further permissions regatding the use of this work. Publisher contact information may be obtsined at bbnp:/seww stor org/joumalsfuepress hel. The Journal of Policical Economy ©1988 The University of Chicago Press ISTOR and the JSTOR logo are trademarks of JSTOR, and are Registered in the U.S. Patent and Trademark Office. For more information on JSTOR contact jstor-info@unich.edu, ©2002 JSTOR, hup:thrwwjstor orgy Fai Feb 15 12:55:28 2002Gibson's Paradox and the Gold Standard Robert 8. Barsky asec of Michigan and Nosanal Buren of Ecanamic Reworch Lawrence H. Summers Harvard University and Natal Barca of Econo: Reseaech ‘This paper comiribuces 2 new clement to the explanation of the Gibson paradox, the puzzling correlation between interest rates and the price level seen ding che god standard period, A shack tha raises the underlying ceal rate of retuen in the economy reduces the equilibrium relative price of gold and, with the nominal price of gold pegged by the authorities, must raise the price level. The mechanism {nwoives che allocation of gold between mionetary and nonmonetary uses, Our explanation helps co resolve same important anomalies ia previous work and is supparted by empirical evidence along a num- bee of dimensions, Monetary theory leads us to expect a correlation between nominal interest rates and the race of change, rather than the level, of prices Yet, as emphasized by Keynes (1936), vo centuries of data do not ‘confirm this expectation. Between 1730 and 1980, the British consol yield exhibited close comovement with the wholesale price in- dex, alongside an essentially zera correlation with the inflation rate Keynes referved to the strong positive correlation between nominal We are grateful to Andy Abel, Rober: Barro, Olivier Blanchard, Brad DeLong, Suanley Fischec, Mikon Friedman, Greg Manki, Franco Modight, Julio Rotember, ‘Anna Schwartz, Robers Shiler, particpanca in seninare t Columbia, Harvaedy Mary land, Michigan, Massachusetis Tnytitute of Technology, University af California, Los Angeles, National Bureau of Economic Research, and the Federal Reserve, and several anoaymous referees foe helpful caraments “This wark is an ourgrowih of Barsky's fastens thesis a MET, completed 1984, ovat of Fatal aan, 0 Sin ey ib al i eee C0 604A 609.00180 0 58)‘canson’s PaRaDox 539 interest rates and the price level, which he called Gibson's paradox, as “one of the most completely established empirical facts in the whole field of quantitative economics” (Keynes 1930, 2:198}, Fisher wrote that “no problem in economics has been more hotly debated” (Fisher 1936, p. 399} Fisher attempted to resolve the Gibson paradox by combining his relation between nominal rates and expected inflation with the hy pothesis that inflationary expectations were formed as a long distrib tuted lag on past inflation, His explanation has been widely chal- lenged, Sargent (1973) noted that such a distributed lag appears incompatible with the stochastic process actually followed by inflation im the Gibson paradox period. Shiller and Siegel (1977) reported that movements in nominal rates during that period seer to be attribut- able to variation in real rates rather than the inflation premium, Keynes (1930) and Wicksell (1936) argued that shifts in the profitability of capital would be accompanied by accommodative movements in the stock of inside and outside money through the behavior af private and central banks. The Keynes Wicksell approach founders on the observation that “neither changes in banks’ reserve ratios nor in the ratio of the domestic gold stock to high-powered money account for any sizable par« of the long-run movements in the U.S. money stock before 1914" (Gagan 1965, p. 255). Instead, the dominant proximate determinant af the mavement of prices and money during the petiad was variation in the stock of monetary gold. As Friedman and Schwartz (1976, p. 288) conclude, “the Gibsonian Paradox remains an empirical phenomenon without a theoretical ex- planation.” This paper contributes 2 new element to the explanation of the Gibson paradox. Noting the coincidence of the observation of the Gibson paradox and the gold standard period, we point out that the Gibson correlation may arise as a natural concomitant of a mane- tary standard based on a durable commodity.' Our theoretical expla- ation revalves around the essential aauure of a metallic standard Since the authorities peg the nominal price of gold at a constant, the general price level is the reciprocal of the price of gold in terms of goods, Determination of the general price level then amounts to the microeconomic, problem of determining the relative price of gold. 1¢ gold is 2 durable asset, its price is sensitive to the long-term Following the treatment of the gold standard by Friedman (1953), " Am independen: and contemporaneous contribution that alsa stresses dhe mec ism of the gold sandard 1 Lee and Petiuzn (1086). We point out the essential Gierences between that paper and the present ane below.530 JOURNAL OF POLITICAL ECONOMY we focus on the demand for gold in its real, as well as its monetary, uses. Using a model similar to that of Barro (1979), we show that i innovations in the productivity of eapital are an important exogenous disturbance, as in Wieksell and Keynes, the negative equilibriurs rela- tionship between the relative price of gold and the real interest rate can give tise to Gibson's paradox. Our mechanism, which relies on the allocation of gold hetween monetary and nonmonetary uses, is (unlike the Keynes-Wicksell mechanism) consistent with the stylized fact that prices varied closely with the monetary gold stock during the gold standard period, The paper is organized as Fallows. Section I refutes recent claims by Benjamin and Kochin (1984) that much of the Gibson correlation is spurious and demonstrates the temporal coincidence of the Gibson contelation and the gold standard, We also present evidence from equity yields indicating that the Gibson correlation held for real as well as nominal assets, Section [I presents a theory of the price level under a gold standard and shows how the Gibson correlation arises naturally in such a setting. Section JII shows that the inverse relation- ship between the relative price of gold and the real interest rate, which provides the basis for our resolution of Gibson's paradox, is a dominant feature of actual gold price fluctuations in the post-1970 period, when the nominal price of gold has floated freely. Section LV provides some limited evidence that two key ingredients of our the- ‘ory, productivity shocks and substitution between monetary and non monetary gold, were in Fact imporcanc features of che gold standard period. Section V contains a brief summary and conclusions. I. Gibson's Paradox in World Data, 1730-1938 This section examines world data on commodity prices, long-term interest rates, and stock yields in an effort to characterize Gibson's paradox. We address the arguments of Benjamin and Kochin (1984) concerning the spurious regression problem and go on to show that Gibson's paradox was primarily a gold standard phenomenon. Then, using stock yield data, we argue that Gibson's paradox involved the underlying real rate of return, and not merely the nominal yield on nominal assets We work with both British wholesale prices? and. a world price index, which is a GNP-weighted average af the wholesale prices of * The Brith price data are from Mitchell and Deane (1962) and were assembied by linking che Elizabeth Schumpecer index with she anqual average ofthe Gaver, Roxon, and Schwartz manthly sndex of Brash camsenodity priges and then (begining 1846) fhe Sauerheck- Statist Overall price indexGinsoy's PARADOX Bat four countries.’ In fact, the correlation of the British price series and ‘our world price level is 96, and very similar results are obcained using cither index. We take the yield on British consols to be the world long- term interest rate.* Was There a Gibson Paradax? Data on world prices and interest rates for the period 1821-1913 are plotted in figure 1, We use the consol yield because, as discussed below, our theoretical model points mast clearly toward a relationship becween long-term interest rates and gold prices. In table 2 below. we also present results using the short-term interest rate. Although a lear positive relationship is observable, Benjamin and Kachin (1484) note that the two series are very nearly random walks, and thus the tisk of spurious correlation is high. Granger and Newbold (1974, 1977) show that an ordinary (testis very likely to show a “significance” relationship between two random walks, even if they were generated independently. These authors also show that standard procedures for correcting serial correlation are inadequate when the errar process involves @ unit root. We deal with the spurious regression problems in two ways. First, Wwe run the regtession in first differences, a standard diagnostic proce- dure recommended by Granger and Newbold (1977) and Plosser and Schwert (1978). Because first-differencing accentuates the high fre- quency variation in the data at the expense of the low frequencies (Anderson 1971) and because it may exacerbate the problem of er- rors-in-variables (Plosser and Schwert 1978), we also report regres: sions in the levels of prices and interest rates. Phe simulation studies of Granger and Newbold (1974, 1977) provide some rough guidance as to the correct critical levels for rejection of the aull hypathesis that ‘wo random walks are independent. They suggest that an ordinary t- statistic greater than 10 or so (corresponding, with 50 observations, to $ We construct the world price index because out model, like that of Barro (979). concerts the warld price level under 4 gold tandard. ‘The four countries are Briain, France, Germany, znd the United Scares although we exclude the US data during the ‘Gred War period. "The weights are feam Baitoch (1982), who atemps to proxy manu coring autput of 2 number of countries in the years 1960 atl 1813. The prices Cor France and Geemany ace front Mitchel (1879), while those for the United States are fram Warren and Peardon (1983), * Following the suggestion of Kamer (1927) and Shiler and Siegel (1977), we use che yields on 2.3 percent government annuties far che years UBBI—BS tsiead of coml elds. During this period, yields had fallen below the 3 percent rate at which consols Iuere ised, and the possbiicy oF government redemption (which actully cecureed the "refunding of 1888") kept the yields on eonsols fom falling euch Further.532 JOURNAL OF POLITICAL ECONOMY 45 — warts Price == Consol Yiele 7620 1840 1860 saa0) 1900 Fic. L-—The world price level acd the consal yield an R¥ of about .7) would properly lead to rejection at the 5-10 per- cent level, We take this as our criterion for significance of the regres: sions in levels ‘Table 1 presents Gibson regressions, in both levels and first differ- ‘ences, for various subperiods of 1730-1938 (able 2 reports analo- {gous regressions using the British open market rate of discount, 2 short-term rate given by Homer [1977]}. The results justify several importanc conclusions. First, Gibson's paradox is not an example of the spurious regression phenomenon, at least not during the classical gold standard years 1821-1913, The regressions in table | using the ‘consol yield in differences are significant at the 1 percent level for the period as a whole and at least at the 5 percent level for both of the subperiods 1821-71 and 1872-1913, The regressions in levels for the whole period have éstatistics in excess of 10. The results in table 2 using the short rate are somewhat weaker but consistent. Second, Gibson’s paradox is by no means a wartime phenomenon.TABLE 1 Recressiox oF Locasrrsn oF Paice LEVEL oN CaNSOL RATE Levels and Fire Differences} Coefficients Durbin Price LevelvFir of Wason Sampic Period Series Differences Consol Yield Stauste 1739-96 Bridsh Levels is “8 (0) First ditferences oe bee ot 03} 1707-1820 rsh Levels — 08 mo 054 Firat differences 05 beh 06 cou 1821-1913 World Levels 0 om co34 First differences 13 vst cay Bish Levels 38 Ae coy First differences 16 mo 05) gar World Levels a7 eae Fiese differences mow British Levels so Fine diferences 7a He72-1913 World —_Levets aa First differences mea Beith Levels 2 First differences iso 1872-1938" Bris Levels 408 Fir differences ne) to2t-age Briksh Levels ae Fira differences Ls? oe Tio pie sera OU534 JOURNAL OF POLITICAL ECONOMY TABLE 2 Rewrsston oF Locaerrnw oF Price Levet 0% Baris Oven Masuer Rave oF Discouwe (Levels aca First Differences) ee Coefficients Durbin Sample Penod Levelstiese of wason an Price Series Differences Conssl Yield Stxtite Iaza—1919) World Levels 7 35 2 cou Free aitferences ol? Lea 15 094) Beis Levels as 36 35 con, First aitferences 025 18 3 oo e26-71 ‘Wor Levels 93 38 2 cay Fess dutferences 2 Lez 20 cou) Beis Levels on 50 ea cay isc differences 0 189 st ot) 1872-1913: World Levels a 20 u (03) Fist differences one Les 2 (093) British Level o 19 ea con Fir differences of 76 99 con Levels 25 a Fest aitferences 30 2 100) 1991-38 Brush Lees to 8 St cos) Firs: differences 05 128 10 03) [Not only is the relation significant and stable during the peacetime, gold standard years from the 1821 resumption of the gold standard in Britain ta the eve of World War 1, but it completely breaks down, during the Napoleonic war period of 1797-1820, when the gold stan- ard was abandoned, Over this period, the correlation was negative in both levels and first differences. These findings are evidence againstGInSON’S PARADOX 535, theories based on government purchases (Benjamin and Kochin 1984) or finance (Keynes 1930; Shiller and Siegel 1977) during war- time. Finally, the evidence of the Gibson correlation is weaker for the pre-Napoleonic period 1730-96 and the interwar years 1921-38 than for the classical gold standard period. Icis plausible to relate the weak, but not nonexistent, evidence of the Gibson paradox from these periods to the rather restricted functioning of the gold standard during them, Although Britain was effectively on a gold standard between 1730 and 1796, most of the rest of the world was nat. Like- wise, the post-1921 gold standard was closely “managed” by central banks and encumbered by formal and informal restrictions limiting convertibibity. ts There Sill a Gibson Paradox? An important question, and a frequent source of confusion, is ‘whether or noc Gibson's paradox persists into the post-World War IT period. Some authors have concluded that it does, on the basis of raw correlations in levels. This is inappropriate for a period during which the price level rose in every year. To establish an economically mean- ingful Gibson paradox, one would aced to show that when the rate of inflation slowed (remaining positive, hawever), the interest rate con- tinued to rise with the price level. That this was not the case is clearly seen in figure 2.° As becomes especially clear after 1965, the interest rate follows the rate of inflation rather than the price level. The complete disappearance of Gibson's paradox by the early 1970s coin- cides with the final break with gold at that rime. The results of this and the preceding subsection corroborate the view of Friedman and Schwartz (1982) that Gibson's paradox is largely, ar perhaps solely, a gold standard phenomenon: “For the period our data cover [1880-1976] i [Gibson's paradox} holds clearly and unambiguously for the United States and the United Kingdom only for the period from 1880 to 1914, and less cleatly for the incer- war period” (p. 586). * Robert Barco, in private communication, informed us that a positive relationship beeen tempoaiy greene porns ad ante toe ha rage nineteenth century but 1 relationship between government purchases and the price level olds ony for pends of suspension. r e “igure 2, which stows the Semonih Treasury bill rate alongsige the level of che ‘Gonguner Price Tadex (CPU) and a month moning average of ination, extends 3 Sima chant presented in Friedman and Schwartz (1976).