Are Technology Improvements Contractionary?

Following Hall (1990), the paper relates each industry’s output growth to a cost-share-weighted sum of input growth plus a residual, but decomposes that residual into observed input growth, unobserved growth in utilization (capital’s workweek and labor effort), and “pure” technology growth, writing dy_i = γ_i(dx_i + du_i) + dz_i, where du_i = s_Ki·da_i + s_Li·de_i captures capital utilization (a) and labor effort (e) using the cost-minimizing firm’s observable proxies for these unobserved margins. “Purified” aggregate technology change is then an appropriately weighted sum of these industry-level dz_i residuals – correcting, in the authors’ words, for “non-constant returns and imperfect competition, and aggregation effects,” on top of the utilization correction. The paper reports that purified technology “varies about half as much as TFP”: its standard deviation for the entire private non-mining economy is 1.5% per year, versus 2% for the raw Solow residual (TFP), so technology’s variance is “only 55 percent as high” as TFP’s variance.

The paper explains that, contemporaneously, “purified” technology is “significantly negatively correlated with inputs” and “uncorrelated with output” – the opposite of the well-known procyclicality of the raw Solow residual (TFP) – because “when technology improves, unobserved utilization as well as observed inputs fall sharply on impact,” so “TFP initially rises less than technology does.” The paper argues this resolves an identification puzzle: if technology shocks were the only disturbance, and they are themselves negatively correlated with the cycle as the paper’s estimates imply, TFP would look countercyclical even without any utilization mismeasurement; what actually makes observed TFP procyclical is that demand shocks (not fully controlled for here) raise output, inputs, and unobserved utilization together, so “changes in utilization make observed TFP procyclical” even though “purified” technology itself is not.

The paper reports that technology improvements “reduce hours worked within the year, but increase hours with a lag of up to two years,” while “output changes little on impact, but increases strongly thereafter”; impulse responses suggest that “at longer horizons… output rises about 1-1/2 times as much as technology.” Non-residential investment “falls sharply on impact before rising above its steady-state level,” while household spending – “especially durable consumption and residential investment” – “rises on impact and rises further with a lag.” The authors summarize that “after a year or so, the response to our estimated technology series more or less matches the predictions of the standard, frictionless RBC model. But the short-run effects do not.”

The paper reports that the price level for non-farm business “falls about half as much as the technology improvement on impact,” with a cumulative decline of “about 1 percent,” while the nominal federal funds rate and 3-month Treasury rate “decline noticeably and remain below normal for an extended period,” and the real interest rate “appears to decline, but modestly.” The relative price of investment to consumption goods barely moves, which the authors read as evidence that, on average, “technical change… is largely neutral” rather than investment-specific. This price-and-interest-rate pattern – a short-run fall in prices accompanying a fall in measured input use – is exactly what a sticky-price model predicts: a technology improvement raises full-employment output and so creates deflationary pressure, which “gives the monetary authority room to lower interest rates,” but if the central bank reacts only with a lag, input use and investment demand still fall in the short run even though prices have begun to adjust.

Using variance decompositions from the paper’s VAR impulse responses, “at the business-cycle frequency of three years, technology shocks account for more than 40 percent of the variance of output, but only 9 to 17 percent of the variance of different input measures.” The authors describe this pattern as “intuitively sensible”: since technology improvements have a near-immediate, close-to-one-for-one effect on output in the long run but depress inputs in the short run, a shock that matters a great deal for output’s long-run trend need not explain much of the (more transitory) cyclical variance in hours, employment, or utilization.

The paper emphasizes that its production-function approach relies on “completely different assumptions for identification” than Galí’s (1999) long-run-restriction SVAR approach – it does not depend on a theoretically derived long-run restriction that might be violated by increasing returns, permanent sectoral shifts, or capital taxes, and it allows examining many variables “without having to modify [the] basic identification strategy.” Despite these different assumptions, the paper notes that “state-of-the-art versions of four very different approaches give similar results”: their own production-function measure; Galí’s (1999) SVAR and its successors; estimated structural DSGE models (citing Smets and Wouters 2003 and Galí and Rabanal 2004); and Shea’s (1998) technology proxy built from R&D spending and patent activity innovations. The authors conclude “it thus appears we have uncovered a robust stylized fact: technology improvements are contractionary on impact,” which “models need to explain.”

The paper directly addresses this concern (raised against an earlier draft) and finds that neither Jorgenson’s nor the BLS non-farm business hours series “is even remotely significant” in Granger-causing purified technology (p-value of 0.35 in levels), and that CEV’s own preferred private-business-hours measure is only marginally significant (p = 0.11). The paper traces the modest significance that does appear to a low-frequency correlation with the productivity slowdown rather than genuine short-run endogeneity: farm employment, a poor proxy for farm hours but a strong proxy for the mid-century productivity slowdown (falling by more than half from 1949 to 1971 before leveling off), does Granger-cause purified technology at under a 2% p-value, but when the specification excludes the trend break that captures the slowdown, the Granger-causality significance for BLS private business hours rises to an insignificant p = 0.39 – leading the authors to conclude “CEV’s Granger causality evidence reflects a low-frequency correlation, not high frequency ‘measurement error’ in purified technology.”

The paper reports that Marchetti and Nucci (2004), applying “exactly our identification method to Italian firm-level data,” find – using additional firm-level data on how frequently individual firms change prices – that “technology improvements reduce input use only at the firms that have sticky prices.” The authors call this “strong, direct evidence in favor of our preferred interpretation,” since it discriminates between the sticky-price explanation and several competing stories the paper separately entertains and only partially rules out: “real inflexibilities” in aggregate demand, sectoral-shifts models, and “cleansing effects” models in which recessions prompt the exit or reorganization of low-productivity firms – while cautioning that “it remains possible that other models could be consistent with the evidence as well” and that these explanations “are not mutually exclusive.”

The paper argues that because its estimated responses are “a composite of the partial effect of a technology improvement and the reactions of policy… to that technology shock,” and because a Federal Reserve that fully offsets a technology shock’s deflationary effect would mechanically erase the measured contraction, the fact that a contraction still shows up in the 1949-96 data implies “monetary policy in the United States over [that] time period did not respond sufficiently to technology shocks to allow actual output to adjust quickly to the new level of full employment output.” The authors connect this directly to the live policy debate (at the time of writing) over whether U.S. technology growth had accelerated in the 1990s, arguing that “short-run movements in technology growth matter just as much for the proper conduct of monetary policy as the long-run rate of technology growth – if not more,” since monetary policy’s central concern is short-run stabilization around a moving full-employment target.

The paper’s conclusion states plainly: “Technical progress is contractionary in the short run, but has its expected expansionary effect in the long run,” and the authors “advance price stickiness as the major reason for the perverse short-run effect of technical improvement,” a view they share with Galí (1999) and Galí and Rabanal (2004). They are careful to note, however, that “our results are not a ’test’ of sticky-price models of business cycles, even though the results are consistent with that interpretation,” and that establishing price stickiness as the unique correct explanation (rather than one contributing factor among several) would require “additional, sharper tests.”