Monetary Policy Lessons of Recent Inflation and Disinflation

Questions & answers

Q1. What lessons is Poole trying to draw, and from what U.S. episode?

Poole surveys the U.S. inflation and disinflation experience of roughly 1975 to 1987 for what it teaches about money demand, real interest rate behavior, market expectations of monetary policy, and the design of monetary rules (Introduction, p. 73). This is explicitly a survey and analytical piece with no formal causal identification strategy: the empirical content is a set of OLS money-demand regressions plus published estimates borrowed from other authors’ event studies, not a structural VAR or narrative-shock design. Poole frames his own position against two alternative readings of the era’s evidence: some economists (he names Benjamin Friedman) concluded the instability of money demand means “policymakers cannot rely at all on any presumed money demand regularities,” whereas Poole’s own “working hypothesis” is that money demand “continues to be a stable function of relatively few variables, but that the interest elasticity of money demand is substantially higher than previously thought” (p. 73).

Q2. What happened to the “standard” money demand function after 1981, and why did it matter?

By 1975 most economists had settled on a money demand specification with income elasticity in the neighborhood of 0.6 to 0.8 and interest elasticity of only about 0.15 to 0.25 (Goldfeld, 1973), a specification consistent with M1 velocity’s steady roughly 3 percent per year trend increase from 1953:1 to 1979:4 (3.1 percent per year, Figure 1). This standard model, together with a two-decade run of rising inflation, supported confident monetarist policy advice through the late 1970s. But M1 velocity departed convincingly from its established trend starting around 1981-82: trend-adjusted velocity fell far more than the actual 1981-1986 decline in interest rates could explain under the conventional elasticity estimates, and Poole reports that he and other economists had explicitly based policy advice in the late 1970s on the joint assumption of a 3 percent velocity trend and an “essentially zero” interest elasticity – an assumption the 1980s data contradicted.

Q3. Why does Poole think first-difference estimates of the interest elasticity are biased toward zero?

Poole identifies two distinct sources of downward bias in the standard first-difference specification. First, a mechanical point: a constant term in a regression run on first differences of the data is mathematically equivalent to a linear time-trend term in the corresponding levels regression, so when the interest rate itself is trending, first-differencing lets the constant “steal” the explanatory power that should belong to the interest-rate variable, understating its coefficient (pp. 78-79). Running Table 2’s first-difference regressions of Δlog(M1 velocity) on Δlog(Aaa bond yield) illustrates the resulting instability: coefficients of 0.238 (1916-86), -0.161 with the wrong sign (1919-46), 0.071 (1947-81), and 0.284 (1916-57), none of it a stable estimate of a single structural elasticity. Second, a substantive “buffer stock” argument (pp. 82-85): in the short run, money balances absorb unanticipated disturbances from spending, credit markets, and income, so the “residuals” of a conventionally specified money demand equation are correlated with credit-market and money-supply disturbances rather than statistically independent, as the standard specification assumes. Because a large share of quarter-to-quarter and year-to-year interest-rate variation reflects these correlated disturbances, treating them as pure demand shocks in a first-difference regression biases the estimated interest elasticity toward zero. Poole is explicit that this buffer-stock mechanism is “qualitative, not quantitative”: he does not estimate it directly, because there is “no obvious way to get around the problem that money demand disturbances are correlated with other variables and/or disturbances in the system” (p. 84).

Q4. What interest elasticity does Poole obtain instead, and how confident is he in it?

Re-estimating log(M1/Y) in levels, with the income elasticity constrained to 1.0 and log(Aaa bond yield) as the interest-rate regressor, over sample periods chosen so the interest rate is approximately equal at both endpoints (1915-1964 and 1920-1968) to minimize the trend-attribution bias described above, Poole obtains a coefficient of about 0.6 in absolute value – roughly two to four times the 0.07-0.28 range from the first-difference regressions – which he explicitly describes as only “tentative” (p. 85, Table 3). He flags a real limitation even for the levels approach: with highly serially correlated residuals, a levels regression can still yield biased coefficients if the residuals themselves follow something like a random walk with drift, which is why he restricts his preferred estimate to the two sub-periods where the interest rate starts and ends near the same level. He also states plainly that he made no attempt to estimate the income elasticity in this paper, because postwar U.S. time-series data suffer from “overwhelming” multicollinearity between income and interest-rate trends, and argues that pinning down the income elasticity will instead require micro survey data or cross-country comparisons (p. 85).

Q5. Why does the long-term bond yield fit money demand better than the commercial paper rate?

In the levels regressions (Table 3), the long-term Aaa corporate bond yield produces a materially larger and better-fitting coefficient than the short-term commercial paper rate across every sample window: roughly 0.65-0.67 for Aaa versus 0.19-0.33 for the commercial paper rate, with higher adjusted R-squared for Aaa in each period (1915-86, 1915-64, 1920-68, 1921-86). Poole’s proposed mechanism is that money holders respond to permanent, not transitory, changes in the return on alternative assets: economizing on cash balances requires durable investments (setting up cash-management systems, opening new bank branches to reduce “shoe-leather” costs) that are only worth undertaking if a rate change is expected to be long-lasting, and the long-term rate – as a weighted average of expected future short rates under the expectations theory of the term structure – better proxies that permanent opportunity cost than the short-term rate does. He is careful to note this errors-in-variables story is not automatic: a short rate with higher variance need not by itself produce a smaller money-demand coefficient, and the pattern could in principle run the other way if velocity actually moved with the short rate (pp. 80-82).

Q6. How does Poole interpret the historically high real interest rates of the 1980s?

Real interest rates, which Poole estimates at roughly 1-2 percent in 1953-73 and 0 to -2 percent in 1973-78 depending on the short rate used, rose into a 4-8 percentage-point range in 1980-85 – levels he calls “dramatically higher” than in the preceding several decades (p. 87, footnote 13). The severity of the 1981-82 recession (peak unemployment near 11 percent) is qualitatively consistent with a monetary explanation for the high real rate, but he argues the vigorous 1983-84 growth and continuing, though slower, 1985-87 expansion are not. For 1983-85 specifically, Poole judges that the joint behavior of the real exchange rate (the dollar appreciated more than 60 percent from its 1980 average to its February 1985 high, per Feldstein and Bacchetta 1987), the real economy, and the real interest rate is “simply not consistent with a monetary explanation” (p. 88); he favors the 1981 U.S. tax-law change, which raised the after-tax return on business investment, as the primary real disturbance, while explicitly noting that others attribute the same effect to the federal budget deficit and stating that “for the purpose at hand, the source of the real disturbance is irrelevant.” He also qualifies that the monetary explanation for 1981-82 specifically “may have to be supplemented,” since the real rate then was abnormally high even by the standard of prior postwar recessions, though not by the standard of 1929-32. As an illustration of how large an interest-elasticity-of-0.6 money demand implies real disturbances can be, he calculates that a real-rate rise from 4 to 8 percent, doubling nominal rates (a log change of 0.7), implies a 0.42 log-point (about 50 percent) velocity swing – which he calls “very large.”

Q7. What does market behavior around weekly money announcements reveal about Fed operating procedures?

Using published estimates from Roley and Troll (1983) and Roley (1986), Poole reports that the Treasury bill rate’s response to an unexpected $1 billion weekly M1 surprise rose from a “trivial” 1.6 basis points under the Fed’s October 1977-October 1979 procedure (best described as an adjustable federal-funds-rate peg) to 10.4 basis points under its October 1979-October 1982 nonborrowed-reserves targeting procedure, then fell to 3.4 basis points (October 1982-February 1984) and 1.4 basis points (February 1984-September 1985), with Poole’s own “casual impression” that the response had fallen essentially to zero by 1986 (pp. 90-91). A parallel discount-rate channel moved the same way: per 100 basis points of discount-rate change, the bill rate moved about 10 basis points in the two years before October 1979 versus about 54 basis points in the three years after (Roley and Troll 1983, p. 91-92). Poole’s interpretation is that the informational content of money-stock data to markets is endogenous to the Fed’s chosen operating procedure – highly informative when the Fed was plausibly targeting nonborrowed reserves, uninformative once it returned to interest-rate control – and he cites Engel and Frenkel’s (1984) finding that the dollar appreciated (depreciated) against the German mark on unexpected increases (decreases) in the U.S. money stock over October 1979-August 1981 as evidence that the market’s interest-rate response reflects a real-rate reaction to anticipated policy, not merely revised inflation expectations. He also raises, as his own hypothesis rather than an estimated result, that the Fed’s response to news may be asymmetric depending on whether inflation or recession is the public’s dominant concern (p. 92).

Q8. What does this experience imply for monetary-rule design?

Poole concludes that the steady-state case for a rule of constant money growth is essentially unaffected by 1980s experience, but that the alternative “gradualist” disinflation prescription – a pre-announced, stepwise reduction in money growth to bring inflation down – is “unreliable” (pp. 96-98), a reversal of the gradualist position Poole says he himself held in the late 1970s. His reasoning is explicitly hedged as “very casual” and dependent “almost entirely” on his own “feel for the data”: since the 1981-1986 velocity decline was far larger than any standard model would have predicted, a gradualist money-growth path launched in 1980 (say, cutting money growth by one percentage point per year) would very likely have produced falling nominal GNP – deflation rather than the disinflation actually achieved – and, by his illustrative comparison, could have implied a larger inflation deceleration than that of 1928-32 (a 9.9-point drop in his gradualism scenario versus 9.0 points actually observed 1928-32). He further argues there is a “serious and probably insurmountable” credibility problem specific to pre-announced disinflation: it is irrational for markets to believe a lower-inflation promise absent evidence the central bank is actually cutting money growth, so a recession may be practically necessary to demonstrate seriousness – meaning the disinflation-transition problem, unlike the steady-state rule, has no rule-based solution Poole can offer.