Whither News Shocks?

Abstract

Previous articleNext article FreeWhither News Shocks?Robert B. Barsky, Susanto Basu, and Keyoung LeeRobert B. BarskyFederal Reserve Bank of Chicago and NBER Search for more articles by this author , Susanto BasuBoston College and NBER Search for more articles by this author , and Keyoung LeeFederal Reserve Bank of Chicago Search for more articles by this author Federal Reserve Bank of Chicago and NBERBoston College and NBERFederal Reserve Bank of ChicagoPDFPDF PLUSFull Text Add to favoritesDownload CitationTrack CitationsPermissionsReprints Share onFacebookTwitterLinked InRedditEmailQR Code SectionsMoreI. IntroductionIn the last decade or so, the old "Pigouvian" hypothesis that good news about the future may cause business cycle expansions—which in turn are reversed by less favorable news—has again become the subject of an important discussion within macroeconomics, due in large part to the pioneering work of Beaudry and Portier (2004, 2006). The investigation of news shocks has been driven substantially by the growing realization that it is difficult to identify structural shocks that might be responsible for the bulk of business-cycle fluctuations. Identified shocks to technology do not to induce the right business-cycle comovements,1 nor can technology shocks yield anything close the degree of forecastable variation in hours and output seen in real data (Rotemberg and Woodford 1996). Monetary shocks may induce the right comovements, but their variance is too small. The importance of investment-specific technology shocks appears to decline when the shock is asked to explain the relative price of investment.2 Other shocks are often either implausible (e.g., high-frequency labor supply shocks) or difficult to identify without conditioning on a full dynamic stochastic general equilibrium (DSGE) model (e.g., mark-up shocks).News shocks are appealing for several reasons. First, they are a priori plausible sources of fluctuations. Second, they fix some of the unappealing features of technology-driven business cycle models. For example, recessions can occur without technological regress in models with news shocks; downturns can take place even after positive realizations of technology change, as long as the realizations were smaller than expected. Third, news-driven models of cycles do not require large and frequent changes in current fundamentals, since news about the same realization of future technology can change many times. Thus, it is unsurprising that news-driven stories feature in casual explanations of the recession of 2001, and even the Great Recession of 2008. There is considerable appeal to the story that agents' demand for durable goods (including housing) preceding each recession was predicated on highgrowth expectations that—whether or not rational at the time—proved to be excessively optimistic, requiring a reduction in subsequent purchases and (at least in the case of the Great Recession) a concomitant deleveraging process.On the other hand, news-driven models have a fundamental theoretical difficulty of their own. Notably, in standard neoclassical models, news shocks do not produce the correct business cycle comovements. The reasoning follows directly from the insight of Barro and King (1984): without a current increase in the marginal product of labor, consumption and labor supply must move in opposite directions. Good news shocks may raise consumption, but they will then lower investment, and labor input and output. Much theoretical work has been devoted to augmenting neoclassical models so that good news shocks will be expansionary in these models.3This paper focuses on the empirical identification of news shocks and their effects. As we proceed with the empirical work, we attempt to understand and rationalize what we find in the data in terms of "DSGE reasoning" with and without formal DSGE models.As far as the identification issue is concerned, we focus the lion's share of our effort in the area of structural (or semistructural) vector auto regressions (VARs), for two reasons. First, the bulk of the substantive debate about the empirical effects of news shocks has used the methods of VAR analysis. Second, we believe that the substantially (though certainly not completely) nonparametric nature of the VAR approach (in contrast to fully specified DSGE models) maximizes the likelihood that the answers will be driven by the data rather than by model structure. Having identified the properties of news shocks and shown that they are robust to reasonable variations in specification, we turn to interpreting the results through the lens of a fully specified DSGE model using Monte Carlo simulation methods.Before asking whether news shocks cause business cycles, one ought to ask whether news shocks do in fact exist to a nontrivial extent. Somewhat to our surprise, we find that they do. More precisely, we find that an innovation to the expectation of total factor productivity (TFP) at t + 20 quarters that is orthogonal to current TFP explains between 20 and 40% of the variance of TFP at a horizon of five years, and typically in excess of 50% of the variance at a horizon of 10 years. Since we take the information set to be orthogonal to current TFP, the innovation we identify is a news shock by a stringent definition. Thus, it appears that news shocks exist, and are quantitatively important. A second important preliminary observation is that news shocks have the character of anticipated growth shocks. In response to our nonparametrically estimated news shocks, TFP displays a rather smooth increase over a number of quarters. Though the error bands (computed by the method of Sims and Zha, 1999) are regrettably large, there is no positive evidence that the effect of news shocks on the level of TFP reverts even in the long run. This is in contrast to the parametric news processes in the original modern theoretical news shock paper (Beaudry and Portier 2004), versions of which are often employed in DSGE models. These processes typically feature an anticipated jump in productivity at more and more dates in the future. On the other hand, an important early empirical paper on news shocks (Beaudry and Portier 2004) estimates a smooth TFP response to news shocks, which is far more drawn out than the more detailed work this present paper indicates.The use of a battery of forward-looking variables, including consumer confidence, stock prices, and consumption is key to our ability to identify an important role for predictable fluctuations in the level of future TFP. Thus, even though the univariate representation of TFP is best characterized as a random walk, implying the growth in TFP is unpredictable; the same is not true in a multivariate context. The Granger causality from multiple forward-looking variables to TFP makes TFP growth predictable. An analogous point was made in the context of gross domestic product (GDP) by Cochrane (1994) and Rotemberg and Woodford (1996), both of whom emphasized the role of the consumption share of GDP in predicting future output. Shocks to linear combinations of innovations in the forward-looking variables constitute our estimated news shocks. More specifically, according to our conceptualization the empirical analogue of a news shock is the linear combination of reduced-form VAR innovations that constitutes the revision in the optimal forecast of TFP at some fixed future horizon.Turning now to the substantive results that follow from our identification of news shocks, we find that the most statistically significant and robust results concern not the quantity variables that are the usual focus of business cycle analysis, but inflation and the expectations component of consumer confidence. Strikingly, the historical decompositions of our VAR system show that our identified news shocks mimic closely innovations in consumer confidence. Equally robust and of perhaps even greater interest is the finding that inflation declines by 30–40 basis points after a news shock and in a typical specification stays below its preshock level for 10 quarters or more. This is unusual behavior if one thinks of news shocks as "demand shocks"—that is, shocks that raise output but are not technological in nature—but it is rationalizable in at least some variants of the New Keynesian model, which stresses the role of future real marginal costs in the determination of current inflation.We of course go on to ask whether news shocks are plausibly significant drivers of economic fluctuations. As noted earlier, few if any of the shocks that economists have identified so far (at least using data-driven procedures consistent with large classes of interesting models) can simultaneously match the comovement that is a defining characteristic of business cycles and also explain a large fraction of economic fluctuations at frequencies identified with economic fluctuations. Can our identified news shocks fill this gap?While the case is not completely open and shut, the short answer appears to be no. As previously demonstrated by Barsky and Sims (2011), the impact effects of news shocks clearly does not induce the kind of comovement that is characteristic of business cycles. In most of our specifications, we find that consumption rises when there is good news, but investment, consumer durables purchases, and hours worked all fall on impact. As discussed above and in Barsky and Sims (2011), this is exactly what one would predict in neoclassical models where there is a representative consumer with time-separable preference, the case analyzed by Barro and King (1984). In such settings, news shocks have an income effect but no substitution effect, so consumption and work hours must move in opposite directions, as must consumption and investment. Our results concerning the negative impact effect of news on hours, durables, and investment are weaker than those of Barsky and Sims (2011), suggesting that these authors may have overstated their case—though Barsky and Sims also cautioned briefly against an overemphasis on impact effects. In this paper we go much farther in that direction. One might easily imagine that Beaudry and Portier, as well as other advocates of news-driven theories of business cycles, would be quite happy with an empirical model in which the business-cycle expansion is somewhat delayed, perhaps due to adjustment costs or other sources of inertia.Unsatisfactory impact effects of news shocks, however, are only the first stage in the uphill battle that expectations-driven business cycle theory faces in light of the empirical evidence. As such a theory predicts, in subsequent periods consumption, investment, and durables purchases all rise strongly, and in some specifications hours of work do as well. However, unlike Beaudry and Portier (2006), we find that TFP begins rising markedly one or two periods after the news shock. Thus, it is difficult to argue convincingly that the expansion in business-cycle variables following a news shock is due to the effects of news per se rather than to the change in fundamentals that is heralded by a news innovation. The hallmark of a news-driven business-cycle theory must be that quantity variables—particularly hours and investment in consumer and producer durables—rise in advance of a predictable rise in TFP. It is certainly possible to find specifications of the reduced form VAR in which there is a significant element of this phenomenon, and large confidence bands further militate against a definitive rejection of its existence. However, the canonical case appears to be one in which productivity and quantity variables move largely in tandem.How well can a New Keynesian DGSE model explain the delayed response of real variables alongside the immediate negative response of inflation to a news shock? We take up this topic in the last section, constructing a New Keynesian model that is quite standard except for the addition of real wage rigidity. As stressed by Christiano et al. (2010) and Barsky and Sims (2009), expected future productivity improvements, ceteris paribus, lower expected real marginal costs. If the real wage does not rise too sharply (either on impact or over time), inflation will jump down and stay down, potentially for a number of quarters—just as in the data. Whether this happens or not depends on the persistence of the news process, the monetary policy rule (particularly the extent to which the monetary authority observes a good proxy for the natural real interest rate), and the degree of price and wage stickiness.II. Semistructural VAR Identification of News ShocksA. Basic MethodLet us turn to VAR identification. This is, of course, intimately tied up with the question of the extent to which quantitatively important news shocks exist in the first place. If the data support a finding that indeed news shocks do exist, we will be interested in the shape of the impulse responses of standard macroeconomic variables (including the usual real variables as well as inflation and asset prices) to the news shock, as well as to the surprise technology shock. We will also study variance decompositions and historical simulations. At this point, the reader should regard what we are doing as establishing facts regarding news shocks in a relatively nonparametric way, not constrained by a structural model. In a later section of the paper, we will compare the impulse responses from the empirical VAR to those from Monte Carlo simulations of a simple DSGE model that illustrates some important mechanisms underlying the observed effects of news shocks, and turns out to fit a number of the facts rather well.A structural news shock will be defined as an advance signal that agents receive about future productivity. One example of a stochastic process for log TFP, a, that contains a news shock would be:This is the baseline news specification of Christiano et al. (2010). A second example (Barsky and Sims 2009, 2011; Jinnai 2013) is:In this specification productivity has a time-varying trend, and agents observe the innovation in that trend one period in advance. Hence the news shock is ϵ2,t −1. Since the impulse response to the news shocks we identify via vector autoregression typically displays a rather smooth increase in productivity over several quarters, the latter process is closer to what is nonparametrically estimated in the data.How do we identify news shocks via vector autoregressions? For the purposes of this paper, an identified news shock will be a linear combination of reduced-form innovations that (a) predicts future productivity, and (b) is not "excessively" correlated with current productivity (more particularly, with a productivity measure stripped of the endogenous, utilization-driven component). Criterion (a) needs no discussion (except for issues about the forecast horizon)—it is certainly a sine qua non for a news shock—nor is it difficult to see why criterion (b) is critical. Consider the extreme case where productivity is a univariate random walk that is also not Granger-caused by any known variables. In that case, the best forecast of future productivity is current productivity; the innovations in the two are equivalent at all horizons, and a shock to current technology and a shock to the forecast of future technology are one and the same. Such an innovation is no doubt a kind of news (all innovations are), but hardly the kind of advance information about future technology with which the expectations-driven business-cycle literature is concerned.It is well-known that TFP growth is approximately white noise. Since the univariate TFP process cannot predict future TFP, the possibility of identifying news shocks from the VAR arises if and only if there are observable variables that Granger-cause TFP. These variables may or may not be the actual signals seen by the agents in the model, but they serve as indicators of the signals agents observe. They may be asset prices, survey measures of expectations, or macroeconomic variables such as consumption, investment, and hours that reflect—and hopefully reveal—the information possessed by the agents.A positive innovation in the stock price orthogonal to the current productivity innovation but Granger-causing future productivity was the operationalization of a news shock in the pioneering empirical paper of Beaudry and Portier (2006). While there are a number of reasons not to rely too heavily on the stock market as an indicator of technological news (some of them discussed in Barsky and Sims [2011]), we retain the stock price as one indicator of a news shock. Consumer confidence also turns out to be a surprisingly valuable indicator (Barsky and Sims 2009, 2012).As a preliminary exercise (not shown here) we computed impulse responses from four separate bivariate VARs, each involving TFP and one forward-looking variable. In each case, TFP is ordered first. We first reproduced the key result in Beaudry and Portier (2006)—an innovation in stock prices orthogonal to current productivity presages a rather long period of increased productivity growth without any apparent subsequent reversion. Almost precisely the same pattern—with a very similar estimate of the long-run increase in the level of productivity—holds when the forward-looking variable is E5Y, the measure of five-year business expectations from the Michigan Survey of Consumers. Next, inspired by the simple Permanent Income Hypothesis and Cochrane (1994), we considered the implications for productivity growth of the orthogonalized innovation in consumption of nondurables and services. Once again, we observe more or less the same pattern, with a somewhat more rapid step-up in productivity and a slightly smaller long-run effect (though the difference is clearly not statistically significant). Finally, informed by the canonical New Keynesian model in which inflation is a jump variable that presages future real marginal costs, we choose inflation for our forward-looking variable. The result is a mirror image of what we saw in the previous three. Thus stock prices are joined by consumer confidence, consumption, and inflation as forward-looking indicators of future productivity.Barsky and Sims (2012) used an agnostic VAR identification with a medium-sized VAR that includes consumption, GDP, hours, stock prices, consumer confidence, and interest identification all shocks that are orthogonal to the innovation in current productivity, and the shock that explains a of future of productivity. They find that their identified news shocks raise consumption but hours, investment, and GDP in the short run. This finding VAR specifications in their paper, and is consistent with the results of standard neoclassical models. news shocks not to for historical identification in Barsky and Sims is not and there is an from about the to the over which technology shocks are to be it is not that we to a priori that the news shock be with the technology innovation. In this paper, we focus our on the is the expectations and the forecast variable is TFP a of the to standard TFP by Basu, and to a TFP of its component). This is the innovation in the optimal VAR forecast of TFP at some fixed point in the future. We this forecast horizon though we focus on what appears to be the reasonable of five it that we ought to with to the current productivity as is standard in the shock On the one hand, one might that a news by should not current TFP. The current of DSGE for news shocks this that news shocks are shocks that productivity in the to current productivity. On the other hand, a in the real may not have this kind of It is possible that news about future productivity with innovations in productivity technological innovations may be with the that significant to technology will occur in the to or there may be of a general technology In these a structural shock TFP on impact and then it further over In such the n

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 63 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.