WPS6525 Policy Research Working Paper 6525 Explaining the Last Consumption Boom-Bust Cycle in Ireland The Role of News and Noise Shocks Jose Luis Diaz Sanchez The World Bank Development Economics Vice Presidency Operations and Strategy Unit, June 2013 Policy Research Working Paper 6525 Abstract The objective of the paper is to explain the last boom The analysis finds that a large and prolonged disconnect and bust in consumption in Ireland by the failure of between consumption and long-run productivity consumers to correctly distinguish permanent changes in occurred in the years leading to the economic crisis, productivity from temporary changes. It uses a business which led to “over-consumptionâ€? for several quarters. cycle model, where agents update their beliefs about long- A strong downward adjustment in 2008 followed when run productivity using information—that they receive Irish consumers finally realized their mistake. continuously—about the future state of the economy. This paper is a product of the Operations and Strategy Unit, Development Economics Vice Presidnecy. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at jdiazsanchez@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Explaining the last consumption boom-bust cycle in Ireland: The role of news and noise shocks∗ Jose Luis Diaz Sanchez University of Paris 1 Pantheon-Sorbonne Paris School of Economics The World Bank Abstract The objective of the paper is to explain the last boom and bust in con- sumption in Ireland by the failure of consumers to correctly distinguish per- manent changes in productivity from temporary changes. It uses a business cycle model, where agents update their beliefs about long-run productivity using information -that they receive continuously- about the future state of the economy. The analysis ï¬?nds that a large and prolonged disconnect be- tween consumption and long-run productivity occurred in the years leading to the economic crisis, which led to “over-consumptionâ€? for several quar- ters. A strong downward adjustment in 2008 followed when Irish consumers ï¬?nally realized their mistake. JEL classiï¬?cation: E21, E32. Keywords: Business cycles, news shocks, long-run productivity. Sector Board: Financial Sector (FSE). ∗ The views expressed in this article are those of the author and do not necessarily represent those of the World Bank. The author would like to thank Michal Andrle for very helpful comments and Blanchard, L’Huiller, and Lorenzoni for sharing their Dynare code. Author’s e-mail: jdiazsanchez@worldbank.org. 1 1 Introduction Economic theory suggests that since productivity captures the capacity of the economy to produce income, an inï¬?nitely lived agent maximizing inter-temporal utility should base his consumption decisions on future productivity. Conse- quently, lack of understanding of this economic fundamental produces periods of “overâ€? and “under-consumptionâ€?. Because agents cannot observe this macro- variable, they estimate it using not only observed productivity but also using information -that they receive continuously- which can or cannot affect produc- tivity (income) in the future. Since the mid-1990s to roughly the mid-2000s, strong increases in labor produc- tivity were coupled with a sharp rise in Irish consumption. However, since around the mid-2000s, while labor productivity remained roughly stable, consumption not only continued its ascendant path, but experienced even higher growth. In 2008, a sharp adjustment in consumption occurred while labor productivity remained stable. If agents base their consumption decisions only on observed labor produc- tivity, there should not be such instability in the relationship between these two variables. It seems then that information about future changes in productivity has played an important role in consumption decisions in Ireland. The aim of this paper is to explain the last boom and bust in Irish consumption by the idea that a large part of the information about the future received by Irish consumers did not have the expected permanent effects on productivity. In particular, our initial intuition is that because of strong increases in labor productivity from mid-1990s to mid-2005, and because of positive signs received about the future, Irish agents believed that the stabilization of labor productivity that occurred in the mid-2000s was only temporary. Consequently, they “over- estimatedâ€? future productivity, leading to a period where Irish agents consumed at levels above what economic fundamentals (in this case long-run productivity) suggested. When they realized their mistake, they immediately and strongly adjusted consumption expenditures downwards. To give empirical support to our intuition, we built on the business cycle with news shocks proposed by Blanchard et al. (forthcoming ) (henceforth BLL) where agents form their beliefs about future productivity (and hence future income) based on information about the future which sometimes is news, sometimes just noise. Thus, agents face a signal extraction problem that can potentially lead to large disconnects between consumption and economic fundamentals.1 We ï¬?nd that errors in consumers’ beliefs about future productivity were sub- stantial during the last 15 years in Ireland. In particular, Irish consumers’ behavior reacted with an important delay to changes in this variable during the mentioned period. The most important consequence has been the large disconnect between the considered economic fundamental and consumer beliefs since 2005. Indeed, since that date, long-run productivity started a descent trajectory, while agents still were estimating increases in this macro-variable. A strong adjustment in 1 This analysis would not be possible in a typical model with news shocks where agents perfectly anticipate changes in productivity. See Lorenzoni (2011) and Milani (2012) for a survey of the news shocks literature. 2 consumption occurred in 2008 when agents ï¬?nally realized that they were “over- consumingâ€?. Blanchard et al.(forthcoming ) and L’Huiller (2012) ï¬?nd qualitatively the same results as our paper using US data. They both ï¬?nd ï¬?rst a slow learning for consumers about changes in the permanent component of productivity. Second they ï¬?nd an overshooting of their beliefs -on long-run productivity in the ï¬?rst paper, and on permanent productivity in the second- around 2005, followed by a sudden downward adjustment in 2008. The paper continues as follows. Section II presents the basic version of the BBL model. Section III describes the estimation methodology and presents the recovered shocks and states from the estimation. Section IV concludes. 2 The BLL basic model 2.1 The model We consider a framework where productivity is driven by two disturbances: a per- manent and a transitory shock. A key assumption in this model is that consumers do not observe the two shocks separately, but only the realized level of productiv- ity. The permanent shock introduces uncertainty about the economy’s long-run fundamentals. The presence of the transitory shock implies that consumers can- not back out the permanent shock from productivity observations, thus creating a signal extraction problem. We assume that consumers have access to an ad- ditional source of information, as they observe a noisy signal of the permanent component of productivity. This adds a third source of fluctuations, a shock to the error term in the signal, which is called in Lorenzoni (2009) and Blanchard et al. (forthcoming ) a “noise shockâ€?, and which has the features of aggregate de- mand shocks, i.e. the shock increases output only temporarily. Consumers solve their signal extraction problem, form expectations about future productivity, and choose spending based on these expectations. Finally, we assume the existence of nominal rigidities in the economy, which implies that spending determines output in the short-run. We explicitly list below the speciï¬?c assumptions of the model. Productivity at (in logs) is the sum of two components, the permanent xt , and the transitory component zt : at = xt + zt (1) The permanent component follows a unit root process: ∆xt = Ï?x ∆xt−1 + t (2) The transitory component follows a stationary process: zt = Ï?z zt−1 + ηt (3) The coefficients Ï?x and Ï?z are in [0,1[, and t and ηt are independently and identically distributed (i.i.d.) draws from a normal distribution with zero mean and with variances σ 2 and ση 2 . 3 In addition to observing the realized productivity level at each period, con- sumers receive a noisy signal (s) about the permanent component xt : s t = xt + v t (4) where vt are also i.i.d. draws from a normal distribution with zero mean and with 2 variance σv . We assume that consumers set consumption (in logs) ct equal to their long-run productivity expectations: ct = lim Et [at+j ] (5) j →∞ where Et is the expectation conditional on the consumers information at date t, i.e., conditional on current and past values of at and st . We simplify the determination of output by assuming that consumption is the only component of demand, and that output is fully determined by the demand side. Thus, output (in logs) is equal to consumption yt = ct . Finally, we assume a linear production function in labor, so that the labor input nt adjusts to produce yt , given the current productivity at , and then nt = yt - at . We now solve for consumption (ct ). Using (2) we can compute the expected value of cumulative productivity in the long-run: Ï?x lim Et [xt+j − xt ] = Σ∞ j =0 ∆xt+j = E [x 1−Ï?x t t − xt−1 ] j →∞ We replace a with x in (5), since the transitory component disappears in the long-run and, rearranging the equation above, we get consumption: 1 ct = (Et [xt ] − Ï?x Et [xt−1 ]) (6) 1 − Ï?x Finally, to complete the solution of the model, we need to solve the consumers’ signal extraction problem to express the expectations of xt and xt−1 in terms of current and lagged values of the shocks ( t , ηt , vt ). This is done in the next sub-section using standard Kalman ï¬?ltering. 2.2 Solving the consumers’ signal extraction problem We deï¬?ne the state vector Xt = (xt , xt−1 , zt ) , the shock vector Vt = ( t , ηt , vt ) , and the vector of consumers’ observations St = (at , st ) . The notation Xt/t represents the agents’ expectation E [Xt /St , St−1 , ..]. We use the Kalman ï¬?lter to express the agents’ expectations Xt/t in recursive form as: Xt/t = AXt−1/t−1 + K (St − St/t−1 ) = (I − KC )AXt−1/t−1 + KSt , 4 where the matrix of Kalman gains K depends on the variance parameters t , ηt and vt . In particular, when the variances parameters are large, st and at are less precise signals of xt and agents take longer to adjust their expectation xt/t to the true value xt (see Lorenzoni 2009 for more details). Since Xt =AXt−1 + BVt and that St =CXt + DVt , we can re-write the dy- namics of Xt/t as:2 Xt/t = (I − KC )AXt−1/t−1 + KCAXt−1 + (KCB + KD)Vt 3 Results 3.1 Maximum likelihood estimation After having solved in the previous section the consumers’ Kalman ï¬?lter to obtain the dynamics of consumers’ expectations, we proceed to estimate the parameters of the model. The equations describing the equilibrium are the ones included in the vectors Xt , Xt/t , and the equation of the endogenous state (6). The variables observed are ct and at . We use Dynare for maximum likelihood estimation. The consumption series is constructed by dividing real consumption by pop- ulation. The labor productivity series is constructed by dividing employment by real GDP. Both series are taken in logs. An important issue we have to confront is that, contradictory to our model, productivity and consumption have different growth rates over the sample. This difference reflects factors left out of our sim- ple model such as changes in participation rate, or changes in saving rate (likely linked with the deterioration of the Irish current account during most of the stud- ied period), or changes in the capital-output ratio. Consequently, we remove the drift in the consumption-to-productivity ratio applying the ï¬?rst-differencing ï¬?lter to both series. Our dataset spans the period 1997Q1 to 2011Q4. The series for Real GDP, Real Personal Consumption Expenditures, and Employment are from the OECD database. Population data comes from Eurostat and the Central Statistics Office (CSO). Figure 1 plots the data series. 3.2 Recovering shocks and states Figure 2 shows the recovered shocks from our structural estimation. We remark that permanent productivity shocks are much smaller compared to the size of noise and transitory productivity shocks.3 This result highlights the major signal   1 + Ï?x −Ï?x 0 2 From (1) to (4) it is straightforward that A =  1 0 0 ,B =   0 0 Ï?z 1 0 0  0 1 0 1 0 0 0 0 0 ,C = ,D = 1 0 0 0 0 1 0 1 0 3 The high autocorrelation of the permanent shocks is a consequence of the difficulty to identify the precise quarter when the permanent change in productivity occurs. 5 Figure 1: Data Series Note. Index 100 in 1997Q1 for the two data series. extraction problem that Irish consumers faced during the considered period when deciding on their expenditures levels, and that, as we will see below, led to large and prolonged errors in the estimation of long-run productivity. Figure 2: Smoothed estimates of shocks In Figure 3 we plot the estimation of the smoothed series of future produc- tivity (solid line) and the consumers real time beliefs about it (dashed line).4 We can clearly distinguish two different sub-periods. During the ï¬?rst subperiod (mid-90s-2005), with the exception of a few quarters, consumers were system- atically underestimating long-run productivity, while in the second sub-period 4 Equation (6) shows the general expression for expected long-run productivity. 6 (2006-2011) the situation reversed and Irish consumers highly over-estimated this macro-variable. Figure 3: Smoothed estimates of long-run productivity The economic interpretation of our results is as follows. The shifting of re- sources into sectors associated with high productivity growth (chemicals and ICT) (see for instance, Cassidy 2004) led to high increases in long-run productivity from the mid-1990s to mid-2000s in Ireland. Irish consumers seemed surprised with this strong increase and caught up gradually with the high levels of this economic fun- damental. After the mid-2000 the story was reversed. Irish people did not realize immediately the decline on future productivity. The observed previous strong increases in labor productivity coupled with positive signs about the future (that ended up being only noise, as Figure 2 shows), made Irish agents believe that the stop in the rise in labor productivity was only temporal.5 Almost two years after a period of “over-consumptionâ€?, agents ï¬?nally realized this disconnect, and suddenly adjusted their consumption levels downwards in 2008. 4 Conclusion This paper has proposed an explanation of the last boom and bust in consumption in Ireland in terms of a failure of Irish consumers to correctly assess long-run pro- ductivity. Thus, a prolonged “over-estimationâ€? of future productivity was followed by a large adjustment in consumption, more in line with long-run fundamentals. A natural extension of this work is to explore the role of agents’ errors about future productivity in the Irish current account recent fluctuations. Indeed, the current account balance became largely negative since the mid-2000s -curiously a 5 The index of consumers expectations, constructed by the Economic and Social Research Institute (ESRI), shows that the level of the index in 2006 was comparable to the high lev- els observed in the mi-90s, when the imminent creation of the Euro-zone boosted consumer expectations about future revenues. 7 time when productivity stopped increasing- to only recover strongly in 2009 and 2010. References [1] Blanchard, O.J., LHuillier, J.-P., Lorenzoni, G. (forthcoming ). “News, noise, and fluctuations: an empirical explorationâ€?, American Economic Review. [2] Cassidy, M. 2004. “Productivity in Ireland: Trends and Issuesâ€?, Central Bank of Ireland, Quarterly Bulletin, Spring issue. [3] L’Huiller, J.-P. 2012. “Did the US consumer overreact? A test of rational expectationsâ€?, Economic Letters, 116: 207-209. [4] Lorenzoni, G. 2009. “A Theory of Demand Shocksâ€?, American Economic Review, 99(5): 2050 - 84. [5] Lorenzoni, G. 2011. “News and Aggregate Demand Shocksâ€?, Annu. Rev. Eco, 3: 537-57. [6] Milani, F. 2012. “The Modeling of Expectations in Empirical DSGE Models: a Survey â€?, Advances in Econometrics, 28: 1-36. 8