Stock Prices, News, and Economic Fluctuations

The paper’s method is to derive two different orthogonalized moving-average representations of the same reduced-form (TFP, stock price) system: one, ε, obtained by imposing an “impact restriction” that isolates the stock-price innovation contemporaneously orthogonal to TFP; the other, ε̃, obtained by imposing a “long-run restriction” that isolates the shock driving TFP’s permanent movements. The authors show these two schemes make sharply different predictions depending on the true model: in a model where fluctuations are driven by surprise technology shocks that raise productivity on impact (plus, say, a monetary shock), the impact-restricted stock-price innovation and the long-run TFP-driving shock “should be orthogonal” to each other, since they capture logically distinct disturbances; but in a “diffusion” model where forward-looking stock prices price in technological opportunities before those opportunities actually expand the production-possibility set, the two shocks “are predicted to be co-linear,” since both are capturing the same underlying anticipated change, just recovered through different restrictions. The authors label the hypothesized anticipatory disturbance a “news shock… since it is a shock that brings news about future growth in productivity.”

The paper uses quarterly U.S. data from 1948Q1-2000Q4 on a stock-market index (the S&P 500, deflated by the non-farm business GDP deflator and converted to per-capita terms) and a standard TFP measure constructed from BLS labor share, capital services, output, and hours in the non-farm private business sector, finding via cointegration tests (Nyblom and Harvey 2000) that stock prices and TFP are cointegrated and so estimating a vector error-correction model (VECM), with a baseline choice of five lags (validated against a two-lag alternative for robustness).

The paper reports that the correlation coefficient between the two independently-derived shock series, ε₂ and ε̃₁, is 0.97 (standard deviation 0.006) in the baseline VECM specification – and that “these two orthogonalization techniques recover essentially the same shock series.” This finding proves highly robust: estimating the system in levels instead of a VECM raises the correlation to 0.998; using two lags instead of five gives 0.9997; and the authors note this high correlation “appears robust and therefore remains counter to the predictions of simple models with surprise changes in productive capacity.” A direct Granger-causality test confirms that stock prices Granger-cause TFP, rejecting the null of no causality “at the 1%” level.

Using the unadjusted TFP measure, the paper finds productivity “starts growing quickly after the initial increase in stock prices,” peaking after about four quarters – but the authors immediately flag that “our measure of TFP may be an improper measure of technological opportunities since it does not take account of potential changes in rates of factor utilization,” so that “mis-measured TFP responds rapidly due to changes in factor utilization” even if true productivity does not. This motivates the paper’s robustness exercise using the Basu, Fernald, and Kimball (2002) TFP series, which is explicitly “constructed from dis-aggregated data which controls for variable rates of factor utilization” – the same methodology as these authors’ own companion paper, “Are Technology Improvements Contractionary?”

With the BFK utilization-corrected annual TFP series, the correlation between ε₂ and ε̃₁ remains very high (0.989, s.d. 0.025), but “the response of TFP is quite different”: “TFP does not increase for several years” after the stock-price rise, with “the point estimates actually suggest[ing] that TFP only starts growing four years after the initial rise in the stock market.” The authors build an alternative, quarterly utilization-adjusted TFP series (using the BLS capacity-utilization index to adjust capital services) and find a similar delayed pattern – TFP does not overtake its initial level for “approximately 3 or 4 years” – and conclude that, across both utilization-correction methods, “the substantially delayed responses associated with the adjusted measures of productivity constitute the more believable response to the actual changes in technology.”

In a four-variable system (TFP, stock prices, consumption, and hours), the correlation between ε₂ and ε̃₁ remains extremely high (0.993 unadjusted TFP, 0.990 adjusted), and the paper documents “a substantial hump shaped response of hours… last[ing] about 10 to 12 quarters,” with consumption showing a milder echo of the same hump. Critically, using the utilization-adjusted TFP measure, “growth in TFP above its initial level arises only 12 to 15 quarters after the initial jump in stock prices,” so that “the permanent growth in TFP arrives after the period of temporary boom in consumption and hours” – the authors describe the resulting picture as one where “the economy appears to go through an important temporary boom, a slight recession, followed by a period of substantial TFP growth,” which they interpret as consistent with “a period of time where agents in the economy try to best position themselves to take advantage of future technological change.”

The paper’s abstract states that this shock “explains about 50% of business cycle fluctuations,” and the paper’s variance-decomposition results (Figure 19) show that both ε₂ and ε̃₁ “explain a substantial fraction of fluctuations at business cycle frequencies.” The authors explicitly link this to Basu, Fernald, and Kimball’s finding (in the companion paper on technology improvements) “that TFP growth does not have a strong business cycle component,” arguing their own evidence is consistent with that result – since the “main growth in TFP (once corrected for utilization) arises well after the boom period associated with the initial innovation in stock prices” – but goes further by suggesting that “even if surprise changes in productivity may not be an important source of fluctuations, technological change may nevertheless be at the heart of business cycle phenomena through its potential to generate expectational driven booms.”

No – the authors explicitly caution against this inference “for at least two reasons”: first, “our results do not tell us whether stock prices changes in anticipation of future productivity growth are of the ‘right’ magnitude,” so the data are “consistent with stock prices over or under reacting to information”; second, the results describe only average, not case-by-case, behavior, and so are “consistent with the stock market making occasion[al] large prediction errors,” meaning the paper’s findings “should not be seen as ruling out occasional speculative bubbles.”

The paper notes early on that a high correlation between ε₂ and ε̃₁ could also arise “in the case where technological change is embodied in capital and where such embodiment is mis-measured in the capital stock,” since innovations in stock prices would then precede measured TFP growth for purely measurement reasons rather than genuine anticipation. In its discussion section the paper more broadly surveys “the strength and weakness of three different structural interpretations” of its findings – anticipated/news-based diffusion models, models where productivity growth is a delayed byproduct of a period of high investment activity, and other related mechanisms – presenting these as open theoretical challenges for future research rather than claiming to have definitively adjudicated among them.

The paper’s conclusion states that the near-unity correlation it documents between the two independently-identified shocks is “counter to that predicted by simple models where surprise changes in productivity drive fluctuations,” and that explaining the pattern requires “a model… where agents recognize changes in technological opportunities well in advance of their effect on productivity, and where the recognition itself leads to a boom in both consumption and investment which precedes the growth in productivity.” The authors frame this as “highlight[ing] the potential fruitfulness of reexamining the manner in which productivity growth is modelled in business cycle analysis,” explicitly positioning their empirical findings as a challenge for the theoretical literature to address rather than a fully worked-out structural model in its own right.