Revisions in Utilization-Adjusted TFP and Robust Identification of News Shocks

ECONOMISTS have long argued that changes in expectations about future fundamentals are an important source of economic fluctuations. This view has reemerged recently in part due to an influential paper by Beaudry and Portier (2006), who report that news shocks about future productivity are closely related to innovations driving long-run variations in productivity and constitute one of the main drivers of business cycles. While the importance of news shocks for business cycle fluctuations continues to be debated, the main identifying restriction behind news shocks is almost universally accepted: productivity reacts to news shocks only with a delay.¹

In this paper, we critically revisit the zero impact restriction. We argue that popular measures of productivity are likely to be confounded by business cycle fluctuations due to imperfect measurement of factor utilization. As a result, news shock identifications that rely on short-run restrictions, in particular the zero impact restriction, can produce misleading results. We then propose an alternative identification that is robust to cyclical mismeasurement of productivity and apply it to U.S. data.

The starting point of our investigation is the quarterly utilization-adjusted series of total factor productivity (TFP) constructed by Fernald (2014) that has become the main measure of productivity in the news literature. Fernald frequently updates the adjusted TFP series based on new data and, less frequently, implements methodological changes. We document that a switch in detrending methods in the estimation of utilization significantly changes the cyclical properties of this series. The sensitivity of adjusted TFP to a seemingly small change such as this suggests, as Fernald (2014) acknowledged, but otherwise mostly ignored by the literature, that measurement issues with productivity can be quantitatively important.

To assess the consequences of Fernald's revisions for news shock identification, we redo the estimation of Barsky and Sims (2011), which has emerged as one of the most popular identification approaches in the literature. Based on prerevision vintages of the adjusted TFP series, a positive news shock leads to a jump in consumption on impact but an initial decline in hours worked. As a result, the implied conditional correlation of consumption growth with hours growth is negative, leading Barsky and Sims (2011) to conclude that news shocks do not constitute a main driver of business cycles. Based on postrevision vintages constructed with the new estimate of utilization, in contrast, a positive news shock leads to a coincident increase in consumption, hours, and other real aggregates, thereby affording a news-driven interpretation of business cycles as Beaudry and Portier (2006) proposed.

To interpret these results and illustrate the consequences of productivity mismeasurement for news shock identification more generally, we consider a medium-scale New Keynesian business cycle model that allows for multiple sources of unobserved factor utilization. Under certain conditions, Fernald's estimate of utilization coincidences with factor utilization in the model and adjusted TFP provides an almost perfect measure of true productivity. But under alternative yet equally plausible conditions, Fernald's estimate of utilization and therefore productivity is confounded by substantial cyclical mismeasurement. We conduct Monte Carlo simulations to study the quantitative significance of this mismeasurement. The main insight from these simulations is that identifications relying on short-run restrictions and, in particular, the zero impact restriction can be highly sensitive to differences between factor utilization in the model and its estimation by the econometrician. Since factor utilization is not observed directly in the data and different assumptions about factor utilization are difficult to test, a more fruitful approach devises alternative identification restrictions that are robust to cyclical mismeasurement.

In the final section of the paper, we propose such an alternative identification. Building on the premise by Beaudry and Portier (2006) that news shocks capture information about slowly disseminating changes in technology and economic organization that drive long-run productivity, we extract the innovation that accounts for the maximum forecast error variance (FEV) share of adjusted TFP at a long but finite horizon.² This max-share approach, which builds on work by Uhlig (2003), has been used previously by Francis et al. (2014) to identify long-run technology shocks. We differ in that we apply it to adjusted TFP instead of labor productivity and propose it as a possible news identification. Conceptually, the max-share identification is also similar to Barsky and Sims (2011) and many close variants in the news literature, with the crucial difference, however, that it does not impose the zero impact restriction and, more generally, does not rely on short-run fluctuations in productivity. The max-share identification should therefore be more robust to cyclical mismeasurement of productivity, and we verify this through Monte Carlo simulations with our model.³ In these simulations, the max-share identification performs very well as long as mean-reverting surprise technology shocks do not account for a large fraction of the lower-frequency variation in true technology.

Of course, nothing guarantees that the max-share identification captures news shocks as opposed to other shocks driving future productivity. However, when applied to U.S. data, we find compelling evidence in favor of a news interpretation. The shock has no significant impact on adjusted TFP for several quarters but predicts sustained future productivity growth, accounting for 70% or more of TFP fluctuations at long forecast horizons. More important, the shock is associated with large impact responses of two novel indicators of innovation: an index of books published in the fields of technology by Alexopoulos (2011) and an index of technological standardization by Baron and Schmidt (2015), followed by a hump-shaped increase in R&D expenditures and a gradual decline in the relative price of investment goods. Third, the shock generates strong, positive immediate reactions of forward-looking information variables.

In terms of macroeconomic implications, the max-share identification implies very similar impulse responses as the ones that Barsky and Sims (2011) originally reported, with the important difference that all the results are robust to the revisions in Fernald's adjusted TFP series. Consumption increases on impact of the shock and then gradually rises to a new permanent level, while hours worked initially decline and later increase in a hump-shaped pattern before returning to the preshock level. The shock therefore implies a negative correlation between consumption growth and hours worked, which makes it an unlikely source of business cycle fluctuations. Nevertheless, the shock accounts for a large share of macroeconomic fluctuations at medium and longer horizons and generates sharp impact responses of inflation and asset prices.

The fact that the empirical findings from the max-share identification do not substantively differ from those in Barsky and Sims (2011) does not mean that mismeasurement of productivity is without consequence. In particular, there is no free lunch in the sense that because of these measurement issues, one cannot separately identify surprise shocks to current productivity from news shocks about future productivity. This is important for other research that relies on short-run fluctuations in adjusted TFP for identification.

The main lesson of the paper is that cyclical measurement issues can materially affect the identification of news shocks based on short-run restrictions. While measurement error occupies a central role in many fields of economics, it has generally taken a back seat in quantitative macroeconomics. A notable exception is Christiano, Eichenbaum, and Vigfusson (2004), who argue, as we do, that adjusted TFP may be confounded by measurement error.⁴ They then apply the infinite-horizon strategy of Gali (1999) to identify long-run productivity shocks based on the assumption that measurement errors in adjusted TFP are transient. Our paper differs in important aspects from Christiano et al. (2004) and the related literature on long-run productivity shocks. First, the max-share identification proposed here does not impose that technology is the only source of long-run fluctuations in productivity and instead extracts the shock that accounts for the maximum FEV share of adjusted TFP at a long but finite horizon. The max-share approach therefore affords the possibility that other shocks (e.g., a surprise productivity shock) exert long-lasting effects on adjusted TFP and at the same time addresses the criticism that infinite-horizon restrictions imply potentially large biases in finite-order VARs.⁵

Second, the literature on long-run productivity shocks typically uses average labor productivity as the technology measure and is primarily concerned with the dynamics of hours worked in response to a shock.⁶ As such, this literature does not directly relate to the news literature and the idea that improvements in technology disseminate slowly and in a predictable manner. Indeed in many cases—including the max-share implementation by Francis et al. (2014) on which our identification is based—labor productivity jumps immediately because hours fall on impact, thus resulting in capital deepening.⁷ This may lead to the inadvertent conclusion, often imposed in DSGE models, that technology follows a random walk process with only one shock. While our results are consistent with the fact that empirically measured TFP is well characterized by a univariate random walk process, they nevertheless suggest that the permanent component of the series is slow diffusion, which is rather different from the oft-assumed jump process impulse response functions generated from a univariate random walk process.

Within the extensive VAR literature on news shocks, our paper is perhaps most closely related to the one by Barsky et al. (2015). They identify a news shock by imposing a longer-run restriction that is conceptually similar to the max-share approach proposed here but differs in potentially important details.⁸ Using a prerevision vintage of Fernald's adjusted TFP series, they find that their news shock looks quite similar independent of whether they impose the zero impact restriction. Our results confirm their finding in the sense that the initial response of adjusted TFP to the proposed max-share shock is small and insignificantly different from 0 for several quarters. Our contribution relative to the paper by Barsky et al. (2015) and the rest of the news literature is to document the large revisions in Fernald's utilization-adjusted TFP series and show that these revisions can materially affect empirical conclusions about the effects of news shocks based on short-run restrictions.⁹ We propose the max-share approach as an alternative identification of news shocks and show that it is robust to measurement issues, and we go to considerable length to establish the news content of the extracted shock by relating it to measures of technological innovation and forward-looking information variables.

The idea that the slow dissemination of technology implies predictable long-run changes in productivity relates to a recent (non-news) literature on the macroeconomic effects of persistent productivity growth processes. Rotemberg (2003) discusses extensively the available evidence on the slow dissemination of technology and proposes a model in which random technological progress leads to stochastic variations in long-run output while deviations of output from trend are mostly driven by temporary shocks. As in our empirical investigation, he finds that slowly diffusing technical progress leads to a temporary drop in hours worked and economic activity.¹⁰ Lindé (2009) incorporates autocorrelated shocks to the growth rate of productivity into an otherwise standard RBC model. Autocorrelated shocks to productivity growth have the flavor of news, and he shows that incorporating this feature can help reconcile the RBC model with empirical results on the effects of technology shocks on hours worked.

The business cycle literature has typically measured productivity as the residual of aggregate output not accounted for by capital and labor inputs, commonly known as TFP. Economists quickly realized, however, that TFP may be a poor proxy of technology for a variety of reasons, most notably changes in unobserved factor utilization. In response to these concerns, Basu, Fernald, and Kimball (2006) construct an aggregate measure of productivity that takes into account sectoral heterogeneity, imperfect competition, compositional changes in the quality of labor and capital, and unobserved factor utilization. Fernald (2014) extends the analysis of Basu et al. (2006), which is carried out with annual data, to construct a quarterly measure of TFP. Because of the higher frequency, not all of the corrections in the original Basu et al. (2006) series can be implemented, but perhaps the most important one—the adjustment for variable factor utilization—is.

In what follows, we briefly review the construction of Fernald's utilization-adjusted TFP series. We then document how seemingly small changes in the estimation of factor utilization lead to large revisions in utilization-adjusted TFP that materially affect its business cycle properties.

Fernald regularly publishes revised estimates of adjusted TFP based on new data and methodological changes.¹³ Panel a of table 1 reports² key statistics for the vintages of December 2007, December 2013, May 2014, and May 2016, all over the same sample period 1947:3 to 2007:3.¹⁴ The means and standard deviations are similar across vintages. However, there is a marked change in business cycle comovement from the 2014 vintage onward. The correlation coefficient between the 2007 vintage and post-2013 vintages of adjusted TFP growth is less than 0.6. Furthermore, while the 2007 and 2013 vintages of adjusted TFP growth are positively correlated with output growth and uncorrelated with total hours growth, the 2014 and 2016 vintages of adjusted TFP growth are uncorrelated with output growth and negatively correlated with total hours growth.¹⁵

Panels b and c in table 1 decompose these changes in correlation into the parts coming from unadjusted TFP and utilization. The business cycle properties of unadjusted TFP growth remain essentially unchanged across vintages. Variations in utilization, by contrast, are significantly larger for the 2014 and 2016 vintages, and there is an important decline in correlation relative to the 2007 and 2013 vintages.¹⁶

This suggests that the large changes in business cycle properties of adjusted TFP are not due to revisions in nonadjusted TFP (i.e., due to data revisions or changes in NIPA methodology), but are instead driven primarily by revisions in utilization. We confirm this conjecture by combining the 2007 vintage of utilization with nonadjusted TFP from other vintages. Correlations for the resulting synthetic series of adjusted TFP are presented in the online appendix. The correlations of the 2007 adjusted TFP vintage with the synthetic 2014 and 2016 series are both 0.91, compared to 0.56 and 0.58 for the actual vintages. Hence, while revisions in utilization do not account for all of the changes in adjusted TFP, they explain the large majority.

What explains the changes in estimated utilization across vintages? Between December 2013 and May 2014, Fernald implemented two methodological changes. First, he switched from using estimates of industry weights and proportionality factors $β^i$ in equation (4) by Basu et al. (2006) to estimates from Basu et al. (2013), which are based on more recent data and a more detailed industry decomposition. Second, Fernald has to contend with the issue that hours per worker in many industries are trending over time. Up to the December 2013 vintage, Fernald follows Basu et al. (2006) and detrends industry hours per worker with the bandpass filter of Christiano and Fitzgerald (2003) to isolate frequencies between 8 and 32 quarters. From the May 2014 vintage onward, Fernald instead detrends industry hours per worker with the bi-weight filter used in Stock and Watson (2012), which removes a much slower-moving trend than the bandpass filter.

Using replication codes for the December 2013 and May 2014 vintages shared generously by Fernald, we assess the quantitative importance of the two changes. Table 2 reports the results. For comparison, the first and the last columns replicate the business cycle properties of the actual December 2013 and May 2014 vintages of estimated utilization growth from table 1. The second column, labeled $Δlnu^t13,BFFK$⁠, shows the effect of switching to the industry weights and proportionality factors from Basu et al. (2013). While this switch lowers the volatility of utilization somewhat, it leaves the correlation with the 2007 vintage essentially unchanged. As shown by the third column, labeled $Δlnu^t13,BW$⁠, by contrast, changing the detrending method from bandpass filtering to bi-weight filtering leads to a substantial increase in the volatility of utilization growth and a concurrent decrease in the correlation with the 2007 vintage. Finally, as shown in the fourth column, labeled $Δlnu^t13,BFFK&BW$⁠, the two changes essentially replicate the 2014 vintage. The remaining difference is due to data revisions.

The results make clear that the change in filtering of hours per worker is the main driver of the revisions in utilization growth. We wish to emphasize, however, that it is not the bandpass or bi-weight filter per se that matters for the results, but rather the frequencies isolated by the different filters. As noted, the bandpass filter used in the earlier vintages of Fernald's series removes high-frequency fluctuations from utilization, whereas the bi-weight filter does not. It is the inclusion of the higher-frequency fluctuations with the bi-weight filter, not the relatively lower-frequency fluctuations that it captures compared to the bandpass filter, that account for the differences across vintages.¹⁷

For the purpose of the news identification that follows, it is not clear that either the bandpass or the bi-weight filter, or any other statistical detrending method for that matter, adequately captures the appropriate fluctuations in hours per worker to correctly infer factor utilization. This means that for either filtering choice, utilization and therefore adjusted TFP may still be confounded by cyclical mismeasurement even if the conditions underlying the proportionality assumption in equation (4) are satisfied (a point to which we return in section IV).

Starting with Cochrane (1994), the modern macro-literature has defined news shocks as information useful in predicting future fundamentals (often productivity) but unrelated to current and past fundamentals. As proposed by Beaudry and Portier (2006) proposed, this implies a zero impact restriction, which is that news shocks affect productivity only with a delay. This restriction is at the core of almost all news shocks identifications used to date.

In what follows, we use the news identification approach of Barsky and Sims (2011) to quantify the implications of the revisions in adjusted TFP for news shocks, although the lessons learned are relevant for other identification approaches relying on the zero impact restriction as well. We focus on the Barsky-Sims approach because it does not require taking a stand on the nature of non-news shocks.¹⁸ Furthermore, Monte Carlo simulations show that the Barsky-Sims approach performs well provided that productivity is measured correctly. As such, the Barsky-Sims approach has emerged as one of the most commonly used identifications in the literature.

The Barsky-Sims identification consists of estimating a VAR and extracting the innovation that is orthogonal to Fernald's adjusted TFP series but maximally accounts for the FEV share of adjusted TFP over a ten-year horizon. Since our alternative identification proposed in section V is conceptually very similar, we review the details here. Let $Yt$ be a $k×1$ random vector process of which the first variable is a measure of productivity (e.g., Fernald's utilization-adjusted TFP), and let the reduced-form moving average representation of this process be given by $Yt=B(L)ut$⁠, where $ut$ is a $k×1$ vector of prediction errors with variance-covariance matrix $E(utut')=Σu$⁠, and $B(L)=I+B1L+B2L2+⋯$ is a matrix lag polynomial.

Now assume that there exists a linear mapping between the prediction errors and the structural shocks, $ut=Aεt$⁠, where $εt$ is a $k×1$ vector of structural shocks characterized by $E(εtεt')=I$⁠, and $A$ is a $k×k$ matrix satisfying $AA'=Σu$⁠. Given the symmetry of $Σu$⁠, a multitude of $A$ is consistent with $AA'=Σu$⁠. The Choleski decomposition of $Σu$ is one potential solution. Denote this by $A˜$⁠. The entire set of permissible values of $A$ consistent with $AA'=Σu$ is then described by $A˜Q$⁠, where $Q$ is an orthonormal rotation matrix and the structural moving average representation is $Yt=C(L)εt$⁠, where $C(L)=B(L)A˜Q$⁠.

The news shock identification of Barsky and Sims (2011) consists of picking $γ$ to maximize the sum of FEV shares of productivity (the first variable in the VAR) up to some truncation horizon $H$ subject to the restriction that the shock is orthogonal to current productivity.

We apply the Barsky-Sims identification to a four-variable VAR comprising either the 2007 vintage or the 2016 vintage of Fernald's utilization-adjusted TFP series, real personal consumption expenditures per capita, total hours worked per capita in the nonfarm business sector, and inflation as measured by the growth rate of the GDP price deflator.¹⁹ Results for larger VARs that contain additional macro aggregates are similar. With the exception of the inflation rate, the variables enter the VAR in log levels. The VAR is estimated with four lags via Bayesian methods subject to a Minnesota prior.²⁰ Confidence bands are computed by drawing from the resulting posterior distribution. The sample period is fixed at 1960q1 to 2007q3.²¹ As in Barsky and Sims (2011), the truncation horizon is set to $H=40$⁠.

Based on the 2007 vintage of adjusted TFP, the responses are very similar to those estimated by Barsky and Sims (2011). Adjusted TFP starts to increase the quarter after the shock, consumption jumps up while inflation falls on impact, and hours worked initially decline, turning significantly positive only after about twelve quarters. As shown in the appendix, these responses imply that if the economy was buffeted solely by news shocks, the correlation between consumption growth and hours growth would be negative, whereas in the data, this correlation is robustly positive.

Based on the 2016 vintage of adjusted TFP, in contrast, the impulse responses look different in economically meaningful ways. Adjusted TFP reacts to the news shock only after several quarters, while hours worked increase from the beginning (although insignificantly for the first few quarters), reaching peak response about ten quarters earlier than based on the 2007 vintage. This difference in the response of hours worked implies that the correlation of consumption growth and hours growth conditional on news shocks is now positive (see the appendix for details). Moreover, the deflationary impact of news shocks, which Barsky et al. (2015) cite as one of the most robust features of the data, is no longer statistically significant.

The difference in responses depending on the vintage of adjusted TFP used has important implications for the role of news shocks. Based on the 2007 vintage, the absence of comovement between hours and consumption leads Barsky and Sims (2011) to conclude that news shocks about future productivity are not a major source of business cycle fluctuations. Based on the 2016 vintage instead, the coincident increase in consumption and hours is consistent with the view espoused by Beaudry and Portier (2006) that news shocks have significant short-term demand effects and are a potentially important driver of business cycle fluctuations.

The results of the preceding sections illustrate that measurement issues about productivity can have important implications for news shock identifications that rely on short-run restrictions on productivity and, in particular, on the zero impact restriction. To interpret and better understand these results, we build a medium-scale New Keynesian DSGE model and conduct different Monte Carlo simulations to address the following questions. Under what conditions does Fernald's adjusted TFP series appropriately measure technology? What are the consequences of different sources of productivity mismeasurement for news shock identification? Is Fernald's new bi-weight filtered estimate of utilization preferable to the previously used bandpass filtered estimate?

The DSGE model we use is based on Christiano, Eichenbaum, and Evans (2005), Smets and Wouters (2007), and Justiniano, Primiceri, and Tambalotti (2010) but augmented with variable labor effort and hours worked so as to analyze the conditions under which the proportionality result between utilization and hours worked set forth in Basu et al. (2006) holds. The model abstracts from heterogeneity in production across industries and imposes an aggregate production function. While potentially important, these assumptions do not invalidate the measurement issues highlighted here.

To save on space, we focus on the model components that relate to the measurement of technology and utilization. A full description of the model is provided in the appendix.

where $νt$ is a preference shock that evolves according to an AR(1) process; $Ct$ denotes consumption; $b$ the degree of habit formation; $T$ the total time endowment; and $G(ht,et)$ the effective time cost when working $ht$ hours at effort level $et$⁠.

where $zt$ is the intensity with which capital is utilized, $a(zt)$ is a convex adjustment cost to utilizing capital, $Ψ(·)$ is a convex cost of adjusting employment, and $J(·)$ is a convex cost of adjusting investment.²² Physical capital evolves according to a standard law of motion subject to an AR(1) shock to the marginal efficiency of investment (MEI).

Labor services are supplied in a competitive market to a continuum of labor unions $j∈[01]$⁠. Unions transform these inputs into differentiated types of intermediate labor and sell them to a labor packer at nominal wage $Wt(j)$⁠. The labor packer combines the different unions' labor into final labor service $Ls,t$ via a CES technology with elasticity of substitution $εw$ and hires it out to intermediate goods producers at nominal wage rate $Wtl$⁠. Unions are subject to Calvo-style nominal wage rigidity.

Finally, the monetary authority sets the nominal interest rate according to a conventional inertial Taylor-type rule in inflation and output growth.

The calibration of the standard model parameters is based on the estimates in Justiniano et al. (2010) except that we impose a stronger degree of nominal wage rigidity so as to generate model impulse responses for total hours and inflation to a news shock that broadly resembles the ones in the data.²⁵

The autoregressive parameters of the exogenous processes take on standard values, and the volatilites of the shocks are chosen to generate an unconditional standard deviation of output growth of 1% (see the appendix for more details). Consistent with Justiniano et al. (2010), the preference shock and the MEI shock are main drivers of macrofluctuations in the model, accounting together for 55% of unconditional variance of output growth and about 85% of the unconditional variance of total hours growth and consumption growth.

Finally, for the proportionality factor $β^$⁠, we either set it so as to match the variance ratio of true utilization to hours per worker in the model or to $β^=3$⁠, which is approximately the variance ratio of Fernald's aggregate utilization estimate to aggregate hours per worker in the data. This is somewhat lower than the variance ratio of true utilization to hours per worker in the model and thus leads to utilization mismeasurement.

We simulate 10,000 periods of data from the model and assess the consequences of technology mismeasurement. Before doing so, we reemphasize that since the model abstracts from several important features of Fernald's construction of adjusted TFP in the data, the simulations are primarily an illustration of the measurement issues that can arise rather than a full explanation of how Fernald's revisions give rise to the changes in business cycle properties of adjusted TFP that we observe in the data. Nevertheless, we think that these illustrations are quite informative.

Table 3 reports the unconditional correlations between true utilization and estimated utilization and between true technology and adjusted TFP under four scenarios.

The first column shows the case when utilization is measured correctly: capital use is constant (⁠$σz=0$⁠), $β^$ is exactly correct so that the proportionality condition holds, and hours per worker are not filtered. The correlation between true utilization and estimated utilization is 1 by definition in this case, and adjusted TFP comoves very closely with true technology. This suggests that incongruities arising from time-varying markups and nonconstant returns to scale by themselves do not matter quantitatively. The second column shows the case of variable capital use (⁠$σz>0$⁠) and the proportionality factor $β^$ set to 3, but no filtering of hours. Estimated utilization still comoves closely with true utilization, and adjusted TFP remains strongly correlated with technology, albeit less so than when utilization is measured correctly.

The third and fourth columns keep variable capital use (⁠$σz>0$⁠) and $β^=3$ but now also detrend hours per worker with either the bandpass filter or the bi-weight filter. Bandpass filtering clearly imparts substantial additional mismeasurement, with the correlations between true and estimated utilization and between technology and adjusted TFP dropping to around 0.5. In contrast, bi-weight filtering does not affect the comovement of measured utilization and adjusted TFP in any significant way.

It is clear from the table that at least for this particular DSGE model, bi-weight filtering leads to less mismeasurement than bandpass filtering. This is because hours per worker in the model are stationary and the bi-weight filter removes only a very slow-moving trend, which in this case is almost equivalent to no filtering. From the perspective of the model, this is preferable because hours per worker are directly related (although not perfectly proportional) to utilization. The bandpass filter, in contrast, removes high-frequency fluctuations in hours per worker and thus imparts serious mismeasurement on implied utilization. In practice, Fernald must contend with secular trends in industry hours per worker that seem unlikely to be related to utilization. As our model abstracts from secular trends, we are not able to speak to the suitability of either filter in this more general situation, although it is interesting to note that in the data, the utilization series implied by bi-weight filtered industry hours per worker turns out to be quite similar to a utilization series obtained without filtering (see the appendix).

Similar to the data, consumption in the model jumps up on impact of the news shock and then gradually increases further to a new permanently higher level; total hours worked drop slightly on impact and then increase in a hump-shaped manner, and inflation falls sharply on impact and then returns to 0 over the next ten quarters. When the proportionality condition holds and hours per worker are bi-weight filtered, the Barsky-Sims identification performs well in capturing the dynamics to a news shock. The fit is somewhat less close when hours per worker are bandpass filtered, with the responses of consumption, total hours, and inflation displaying mild oscillatory behavior. This is not an issue of the Barsky-Sims identification per se but of bandpass filtering when constructing utilization, which appears to introduce artificial dynamics in the VAR.²⁷ Nevertheless, even with bandpass filtering, the fit with true model responses to a news shock remains good. This provides further confirmation that for reasonable markup variations as implied by our model, the difference between Fernald's construction of TFP and true TFP is quantitatively unimportant.

Second, we consider the case when the proportionality condition does not hold; that is, capital use is nonconstant (⁠$σz>0$⁠) and $β^=3$⁠. As shown in figure 2b, the VAR responses for consumption and inflation come again reasonably close to the ones implied by the model, independent of the filtering method for hours per worker in the construction of utilization. The response of total hours in the VAR, however, now depends significantly on the filtering method. Under bi-weight filtering, total hours slightly increase on impact and remain above the model-implied response for about ten quarters. In contrast, under bandpass filtering, the total hours response is, aside from the initial period, negative for several quarters before increasing in line with what the model implies.

The bottom line of this discussion is that bandpass filtering hours per worker, despite inducing substantial mismeasurement of utilization, can in some cases help smooth out departures from the proportionality assumption in Fernald's proxy of utilization. Bi-weight filtering, in contrast, does not attenuate such departures from proportionality and therefore leaves news shock identifications such as the Barsky-Sims approach that focus on short-run restrictions in adjusted TFP more sensitive to utilization mismeasurement. At the same time, it should be clear that none of our simulation results are general. Indeed, for alternative model calibrations, bandpass filtering of utilization does not attenuate the effects of departures from proportionality and, to the contrary, may in fact exacerbate it.²⁸ Hence, we conclude from these simulations that news shock identifications relying on short-run restrictions and, in particular, the zero impact assumption can be highly sensitive to cyclical mismeasurement of true technology. A more fruitful approach is instead to devise alternative identification restrictions that are robust to cyclical measurement issues. This is what we propose in the next section.

The central idea behind our proposed alternative identification is that new productivity-enhancing technologies disseminate slowly and, if known to agents, constitute news about future productivity growth. As long as productivity in the long run is driven primarily by new technologies, an identification that accounts for most of productivity variations in the long run should therefore capture news. At the same time, as long as this identification does not rely on short-run restrictions and in particular the zero impact restriction, it should be robust to cyclical measurement issues with productivity.

The idea that new technologies diffuse slowly finds ample support in a large microempirical literature: Griliches (1957), Mansfield (1961, 1989), Gort and Klepper (1982), and Rogers (1995). According to Mansfield (1989), for example, the time until half of potential adopters actually adopt a new technology varies between five and fifteen years, depending on technology. While the slow dissemination of new technologies and its implications for the modeling of productivity is discussed extensively by Rotemberg (2003) as well as Comin and Gertler (2006) and Lindé (2009), among others, much of the business cycle literature has modeled productivity as a jump process where innovations lead to an immediate change of productivity to a new level that is either permanent or highly persistent. Yet the assumption of slow dissemination is consistent with the basic insight of Beaudry and Portier (2006) from a bivariate VAR that news shocks identified through the zero impact restriction are closely related to the shocks driving long-run movements in productivity. Our contribution here consists of exploring this insight further by extracting a long-run productivity shock in larger VAR systems, assessing its robustness to the documented revisions in adjusted TFP, and using additional information to interpret the extracted shock as a news shock.

Mechanically, the proposed max-share identification is the same as the technology shock identification of Francis et al. (2014), which in turn builds on earlier work by Uhlig (2003), but differs in propose it as an alternative identification of news shocks and apply it to adjusted TFP instead of labor productivity as the target variable. As we will discuss, this latter difference is important. Because of variations in the ratio of capital to labor—capital deepening–labor productivity responds quite differently to the shock than adjusted TFP, thus making the news interpretation less obvious. Moreover, since capital deepening is endogenous, labor productivity is affected even in the long run by other nontechnology shocks, potentially invalidating identification of technology shocks based on long-run restrictions.³⁰

Compared to other long-run identification schemes employed in the VAR literature, the max-share approach has the advantage of focusing on a long but finite horizon. As Francis et al. (2014) show, this helps to reduce small-sample bias in VARs that, as discussed in section I, can have potentially important effects for infinite-horizon identifications of long-run shocks. In addition, the max-share approach does not impose that technology is the only source of long-run fluctuations in productivity and instead affords the possibility that other shocks (e.g., a surprise productivity shock) exert at least some long-term effect on adjusted TFP.

We apply the proposed max-share identification to the same four-variable VAR as in section III. The horizon at which the FEV share of adjusted TFP is maximized is set to $H=80$ quarters, although similar results would obtain for other long horizons. The estimated impulses responses are reported in figure 1b.

In contrast with the results based on the Barsky and Sims (2011) approach, there is very little difference in the impulse responses between the VAR estimated with the 2007 vintage of adjusted TFP and the VAR estimated with the 2016 vintage. In both cases, consumption jumps on impact and then gradually increases further to a permanently higher level; hours worked decline significantly on impact before turning positive after about five quarters, and inflation drops sharply and significantly on impact of the shock before gradually returning toward its initial level. The only discernible difference is the short-run response of adjusted TFP, which should not be surprising given their difference in cyclical properties. For both vintages, adjusted TFP jumps on impact, although insignificantly so. The 2007 vintage then increases gradually, whereas the 2016 vintage temporarily declines and remains insignificant for more than ten quarters. Both vintages, however, increase gradually at longer horizons and end up two to three times higher than their impact responses. In other words, the max-share shock predicts delayed but sustained future productivity growth. Aside from the adjusted TFP response, these results look close to the original results reported in Barsky and Sims (2011) based on the 2007 vintage of adjusted TFP. Indeed, as shown in the appendix, the median correlation between consumption growth and hours growth implied by the max-share shock is robustly negative, contrary to what we observe unconditionally in the data.

As also shown in the appendix, the responses for consumption, hours, and inflation are robust to replacing adjusted TFP with either unadjusted TFP or labor productivity. There is, however, a sizable difference in the response of these alternative productivity measures to the max-share shock. In particular, consistent with Francis et al. (2014), labor productivity jumps considerably more on impact. This is due to a short-run capital deepening effect: a fall in hours generates an increase in the capital-to-labor ratio, which boosts labor productivity on impact relative to the more gradual increase in adjusted TFP. Failure to take this effect into account may lead to the inadvertent conclusion that technology should be modeled as a random walk process, as is quite frequently assumed in the business cycle literature, which is very different from our finding that the response of adjusted TFP to the max-share shock is insignificant on impact before gradually increasing to a new permanent level that is substantially higher, consistent with the empirical literature cited above that technology is slowly diffusing.

As discussed³ above, the news shock interpretation of the proposed alternative identification rests on several important questions, in particular: (a) Does the max-share shock lead to delayed predictable changes in future TFP? (b) Is the max-share shock correlated with measures of technological innovation?, and (c) Does the max-share shock generate sizable responses in forward-looking news indicators?

For the first question, we already know from the results with the four-variable VAR that the max-share shock leads to persistent and therefore predictable changes in future TFP growth. We now extend the analysis by considering an eight-variable VAR system that contains, in addition to the four variables already included above, real gross domestic product (GDP) per capita, real private investment expenditures per capita, the real S&P 500 index (deflated by the consumer price index), and the Federal Funds rate.³¹ This choice of variables is motivated by the desire to learn about the effects of the max-share shock for other prominent macroeconomic aggregates and by the idea that including forward-looking information variables may help sharpen the results and address issues of nonfundamentalness (Leeper, Walker, & Yang, 2013). Indeed, as Beaudry and Portier (2006) argue, a large literature suggests that stock prices reflect expectations about future economic conditions and should therefore be an important indicator of news. Similarly, the Federal Reserve may have superior forecasting abilities and thus, news could also be reflected in the Federal Funds rate, the main monetary policy instrument up until the recent financial crisis. As before, the VAR is estimated with four lags for the 1960:1–2007:3 period subject to a Minnesota prior.

Figure 3 displays the impulse responses. The estimated responses again match closely across the two vintages, confirming the robustness of the max-share approach to the revisions in Fernald's adjusted TFP series. Compared to the four-variable VAR, the reaction of adjusted TFP to the shock is more delayed and gradual, with an impact response for the 2016 vintage that starts closer to 0. This difference in results is primarily due to the inclusion of the real S&P 500 index in the VAR, confirming the point of Beaudry and Portier (2006) that stock prices contain valuable information about market expectations of future economic conditions.

The real S&P 500 index itself reacts strongly on the impact of the shock and then displays a mild hump-shaped response that is quite persistent. Investment and total hours worked both decline initially, while output rises slightly and consumption jumps up robustly on impact. Thereafter, output, consumption, and investment gradually increase toward a permanently higher level, while total hours worked responds in a hump-shaped manner similar to its response in the four-variable VAR. Inflation and the Federal Funds rate both decline significantly on impact and then remain persistently below their original values. The initial decline of inflation substantially exceeds the decline in the Federal Funds rate, implying that real short interest rates increase on impact of the shock. Hence, the shock triggers a contractionary monetary policy response despite the deflationary effect that the shock has on the economy.

The opposite-signed impact responses of consumption relative to hours and investment imply that the max-share shock generates negative business cycle comovement between these variables. This confirms the conclusion from the four-variable VAR that the shock is unlikely to be a main driver of business cycle dynamics. This does not mean, however, that the shock is unimportant for macroeconomic fluctuations more generally. As we document in the appendix, while the shock accounts for only a small fraction of the FEV of real macroeconomic aggregates at short horizons (with consumption being the notable exception), the shock is the main driver of these variables at longer horizons with the exception of hours worked. Indeed, at the eighty-quarter horizon, the shock accounts for about three-fourths of unpredictable variations in adjusted TFP, GDP, consumption, and investment. Quite strikingly, the shock also accounts for almost half of unpredicted variations in the real S&P 500 index and inflation at forecast horizons of twenty quarters and more, and about one-third of unpredictable variations in the Federal Funds rate at horizons of forty quarters or more.

To answer the second and third questions above, we reestimate the eight-variable VAR with the Federal Funds rate replaced sequentially with different measures of technological innovation and forward-looking information variables. The rest of the VAR specification is kept unchanged except when we have to adapt the sample due to data availability, as described below. To save on space, we only report impulse responses for the variables that replace the Federal Funds rate. The seven other variables in the VAR, which are kept the same throughout the exercise, react very similarly to the max-share shock as reported in figure 3.

We first consider four measures of technological innovation: the index of information and communications technology (ICT) standards by Baron and Schmidt (2015), the index of new technology manuals by Alexopoulos (2011), real R&D expenditures per capita from the NIPAs, and the inverse of the relative price of investment price from Justiniano et al. (2010). The index by Baron and Schmidt (2015) counts the number of new ICT industry standards per quarter released by standard-setting organizations (SSOs) in the United States.³² As Baron and Schmidt (2015) argue, standardization is an essential step in the introduction and adoption of new technologies. It precedes the implementation of new technologies but presumably provides an important signal about the commercial viability of an innovation and thus future growth opportunities. As such, standardization represents an ideal measure to assess the extent to which our max-share shock captures news. As in Baron and Schmidt (2015), we focus on ICT standards because they have constituted the dominant type of general-purpose technology, although results are robust to using broader industry standards. Alexopoulos's (2011) count of books published in the field of technology provides a complementary measure even though she develops her measure primarily to investigate the role of contemporaneous technology shocks.³³ As she explains in her paper, new book titles in this area “appear precisely when the innovation is first introduced to market, for the very good reason that the whole purpose of publications is to spread the word about the new product or process.” R&D expenditures and the relative investment price are common measures of the quality and/or efficiency of newly produced investment goods. If our max-share shock captures news about future productivity growth, then we would expect both of these measures to react gradually as new technologies are being implemented and start to affect productivity.³⁴

Alexopoulos's book measure is only available at an annual frequency and stops in 1997. We therefore estimate a smaller, annual VAR for this case, containing adjusted TFP, consumption, inflation, and Alexopoulos's book measure. For all the other variables, the impulse responses are estimated with the above described VAR based on quarterly data for the 1960:3–2007:3 sample.

Taken together, the impulse responses indicate that the max-share shock picks up the introduction of new technologies to markets instead of other shocks that endogenously lead to more R&D activity and eventually more innovation and higher productivity. Otherwise, one would expect ICT standards and new technology book titles to respond not with an initial jump but only gradually and with a delay relative to R&D expenditures.

Next, we consider forward-looking variables that have been interpreted as capturing news: the spread between long-term (five-year) Treasury bond yields and the Federal Funds rate as used in Kurmann and Otrok (2013); the Michigan Survey's five-year-ahead consumer confidence index as used in Barsky and Sims (2012); and the business confidence index from the Business Outlook Survey (BOS) conducted by the Federal Reserve Bank of Philadelphia as used in Bachmann, Elstner, and Sims (2013). Figure 4b shows the impulse responses of these series. For reference, we also include the impulse response of the real S&P 500 index, which is part of the VAR used to generate these results. All of the indicators jump up sharply on impact of the news shock and then decline gradually back to their original level. These responses are highly significant and indicate that the identified max-share shock captures news about the future that is picked up not only by financial markets but also the Fed, consumers, and businesses.³⁵

The results provide compelling evidence that the max-share shock captures news about future productivity growth. The shock predicts delayed sustained future TFP growth, accounting for only a small fraction of TFP fluctuations at short forecast horizons but for 70% or more of TFP fluctuations at longer horizons. Perhaps more important, the shock is associated with large and persistent jumps in two novel measures of innovation, followed by a hump-shaped increase in R&D expenditures and a gradual decline in the relative price of investment goods, and the shock generates jumps in a wide variety of forward-looking information variables. Taken together, these responses suggest that the max-share identification picks up technological innovation as opposed to other business cycle shocks or noise that endogenously lead to changes in productivity and that market participants clearly update their forecasts about the economy. The news interpretation therefore seems natural.

As a final check, we perform the same Monte Carlo simulations as above to assess whether the max-share identification captures more robustly the model responses to a news shock than the Barsky-Sims identification. To save on space, we consider directly the situation where the proportionality condition for utilization does not hold; that is, capital use is variable (⁠$σz>0$⁠) and the factor of proportionality $β^=3$ is different from the variance ratio of true utilization to hours per worker. Results for the case when the proportionality conditions hold are reported in the appendix and match the model responses as closely as the ones obtained with the Barsky-Sims identification.

The Monte Carlo simulation also allows us to assess the robustness of the max-share identification to alternative data-generating scenarios. Generally the max-share identification performs well as long as news shocks account for a large part of the unpredictable variation in adjusted TFP at long horizons (or whatever measure of productivity that one may choose). This performance gradually deteriorates as the importance of other shocks for long-run movements in adjusted TFP is increased, either because these shocks have a direct impact on neutral technology or because of measurement issues. Nevertheless, as we have argued, since long-term changes in productivity are typically slow to diffuse, the assumption that surprise (unanticipated) changes in productivity are important at long horizons seems unlikely. With regard to other shocks that have an impact on adjusted TFP due to measurement error, this is of course a possibility, although one that is true of any identification.

An almost universally imposed restriction in the news literature is that news shocks have an impact on productivity only with a delay. This restriction may be violated if empirical series of productivity systematically mismeasure true technology.

In this paper, we document large revisions in one of the most popular measures of productivity, adjusted TFP by Fernald (2014), and show that these revisions are due to a switch in filtering of hours per worker in the estimation of factor utilization. These changes are evocative of cyclical mismeasurement and materially affect empirical conclusions about the macroeconomic effects of news shock as identified by Barsky and Sims (2011). We therefore propose an alternative identification, based on the max-share approach by Francis et al. (2014), which does not rely on short-run restrictions, in particular, the zero impact restriction. We show that our identification is robust to the revisions in Fernald's series and performs well in Monte Carlo simulations under different assumptions about cyclical mismeasurement of productivity. When applied to U.S. data, we find results that are consistent with a news interpretation: adjusted TFP increases only gradually, whereas indicators of technological innovation and forward-looking information variables jump on impact. The identified shock does not generate comovement in real, and is therefore not a main driver of business cycle fluctuations. This does not imply that the shock is unimportant for macroeconomics as it accounts for the majority of unpredictable fluctuations in real aggregates at medium and long horizons and generates strong impact responses of inflation, the Federal Funds rate, and asset prices. Investigating these results further and assessing the type of models that are consistent with these dynamics are important topics of future research.