Money, Income, Prices, and Interest Rates

The paper asks whether monetary aggregates remain a reliable indicator of future nominal and real U.S. economic activity, and whether the aggregates’ apparent post-1980 predictive failure reflects a genuine structural change in the money-income relationship rather than a sample artifact (Section I, p. 472). This matters because central banks historically used monetary aggregates as information variables for gauging the stance of policy and forecasting income and prices; if that relationship broke down, policymakers need a replacement indicator. The paper’s secondary question is whether an interest-rate variable – specifically the spread between the commercial paper rate and the Treasury bill rate – can serve that role instead (Section II, p. 472).

The analysis uses quarterly U.S. data on nominal GNP, real GNP (1982 dollars), the implicit GNP deflator, four financial aggregates (monetary base, M1, M2, and total domestic nonfinancial credit, all in log first differences), and three interest-rate measures (the commercial paper rate, the Treasury bill rate, and their spread, with the spread entered in levels rather than differenced). The framework is a reduced-form VAR with a uniform four-lag structure, imposed without formal lag-selection tests, and no structural identification is applied anywhere in the paper (Section III, pp. 482-487). Predictive content is assessed with Granger-causality F-tests (joint significance of lagged financial variables in income/price equations) and variance decompositions, run separately for each of the four aggregates. The tests are repeated across three overlapping sample periods – 1960:2-1979:3, 1960:2-1990:4, and 1970:3-1990:4 – specifically to isolate how much the 1980s drive any change in results.

In the pre-Volcker sample (1960:2-1979:3), all four aggregates – base, M1, M2, and credit – are significant predictors of nominal income at the 0.01 level (Table 1); extending the sample to 1990:4 (1960:2-1990:4) causes the base and credit to lose significance entirely, while M1 (3.75) and M2 (4.49**) remain significant at the 0.01 level, and starting the sample in 1970:3 instead (1970:3-1990:4) eliminates significance for every aggregate except a marginally significant M1 (2.27, significant only at the 0.10 level).** Real income results (Table 2) show the same deterioration pattern (Section I, pp. 473-477). Because the same failure shows up across four aggregates constructed very differently (a narrow base, transactions money M1, broad money M2, and total credit), the authors treat this as evidence against a narrow, aggregate-specific explanation and in favor of something more systematic changing across the sample.

The commercial paper rate minus Treasury bill rate spread is significant for real income at the 0.05 level or better in all three sample periods and across specifications, both on its own (Table 2) and when included alongside any of the four monetary aggregates (Table 7); when the spread is present, none of the four aggregates remains significant in any sample (Table 7). In variance-decomposition terms, the spread accounts for roughly 23-32% of real income variance in 1960:2-1979:3 and 21-26% in 1970:3-1990:4 in the base, M1, and credit systems — though the M2-based system shows a notably smaller spread contribution, around 12-15% and 14-22% respectively (Table 8) (Section II, pp. 477-482). The authors interpret the spread as a relative price that reflects credit conditions and monetary-policy stance – widening when risk-averse lenders shift from commercial paper toward Treasuries or when commercial borrowers face higher default premia – and thus as purging common interest-rate variation that a single rate level would still carry (Section II, pp. 477-482, 490-491).

No: tests of the null hypothesis that the paper-rate and spread coefficients exactly cancel (i.e., that the correct specification simply includes the bill rate while excluding the paper rate) are rejected at the 0.05 level in all but one case (Table 9), and a separate test of whether the spread’s own coefficients are all zero is also rejected — “at the 0.05 level or better” for the samples including the 1960s and at the 0.05 or 0.10 level for the 1970-1990 sample. The authors conclude that “what contains statistically significant information about subsequent fluctuations of real income is neither the paper rate nor the bill rate individually, but the relationship between the two” — i.e., the spread’s predictive content is distinct from, not reducible to, any single interest-rate series (Section II, pp. 481-482).

Yes – the cointegration evidence mirrors the Granger-causality deterioration. ADF residual tests (Table 10A) find M2 cointegrated with income at the 0.05 level in 1960:2-1979:3 but find no aggregate cointegrated in 1970:3-1990:4. Johansen bivariate tests (Table 10B) find the base, M1, and credit cointegrated with income pre-Volcker (at the 0.05-0.01 levels) but no aggregate cointegrated post-1970. Trivariate Johansen tests that add an interest rate to the system (Table 10C) find all four aggregates cointegrated in 1960:2-1979:3 and none in 1970:3-1990:4 (Section III, pp. 482-487). Because this is a long-run equilibrium result rather than a short-horizon forecasting result, its consistency with the Granger-causality findings strengthens the case that something structural, not just short-run noise, changed.

In the pre-Volcker sample where cointegration is found, the trivariate cointegrating vectors imply interest semi-elasticities of real money balances of −0.085 for the monetary base, −0.102 for M1, −0.009 for M2, and −0.012 for credit (Table 11), with the base, M1, and credit semi-elasticities statistically significant. These are long-run semi-elasticities from the cointegrating relationship, not short-run dynamic responses (Section III, p. 487).

The paper does not formally identify or adjudicate a structural mechanism; it discusses two candidate explanations without preferring one. The first is that 1980s financial innovation and deregulation (removal of Regulation Q ceilings, spread of NOW accounts, growth of money market mutual funds) disrupted the historical velocity relationship between money and income. The second is that the Federal Reserve’s operating-procedure changes during the 1979-1982 Volcker disinflation severed the earlier money-income link. The authors present both as plausible and explicitly decline to distinguish between them (Section IV, pp. 487-491).

All results are reduced-form: no structural identification is imposed anywhere, and the authors explicitly restrict their claims to predictability rather than causal interpretation (pp. 472, 487-488). The four-quarter lag length is imposed throughout without formal lag-selection tests, and sensitivity to lag length is not reported. The aggregates tested are simple-sum (not Divisia or user-cost-weighted), so the paper cannot say whether the predictive failure is specific to simple-sum aggregation. The spread’s predictive success is documented only through the 1990:4 end of sample, and its stability beyond that period is not examined. The Johansen cointegration tests use standard asymptotic critical values, and possible small-sample size distortions in samples of roughly 120 quarters are not addressed.