Direct and network effects of idiosyncratic TFP shocks

See, for instance, Acemoglu et al. (2012), who argue that microeconomic idiosyncratic shocks can lead to aggregate fluctuations in the presence of intersectoral input–output linkages, thus placing into question the famous diversification argument of Lucas (1977) that microeconomic shocks would average out and are less likely to have a significant impact on aggregate variables. Similarly, di Giovanni et al. (2014) investigated the role of individual firms in international business cycle comovement and showed that direct linkages on comovement at the micro-level has significant macro-implications. For more works in terms of the role of idiosyncratic/microeconomic shocks in macro-fluctuations, see Shea (2002), Horvath (1998), Horvath (2000), Dupor (1999), Carvalho (2008), Acemoglu et al. (2010), Gabaix (2011), Carvalho and Gabaix (2013), Johnson (2014), and Atalay (2017).

Foerster et al. (2011) considered a structural model similar to Long and Plosser (1983) and found that the role of idiosyncratic productivity shocks increased considerably after the mid-1980s, explaining half of the quarterly variation in industrial production.

A summary of AAK’s theoretical model is presented in Sect. 2.

For example: only contemporaneous regressors (\(p=q=0\)), only lags (\(p=q=1\)), and both contemporaneous and lagged variables (\(p=0\) and \(q=1\)).

Real value added is calculated by dividing the nominal value added by the corresponding shipments deflator.

In Sect. 7.2.6, we conduct a robustness check where we exclude the indirect own effects from donwstream measure, thus \(Down_{i,t}= \sum _{j\ne i} l_{ij} Idio_{j,t}\).

In Sect. 7.2.6, we conduct a robustness check where we exclude the indirect own effects from upstream measure, thus \(Up_{i,t}= \sum _{j\ne i} t_{ji}Idio_{j,t}\).

See Foerster et al. (2011)’s Equation 11 and discussions therein.

In Sect. 7.2.3, we provide a discussion when using the industry-level TFP changes as idiosyncratic changes without filtering out the part driven by the aggregate factor.

See Barauskaite and Nguyen (forthcoming) for a discussion about the role of the intersectoral network in propagating aggregate TFP shocks.

“Use” table (U) is a matrix that represents how much each commodity is being used by industries and final consumers. This matrix is presented as a commodities–industries relation. “Make” table (M) is a matrix that represents value of each commodity produces by each industry. This matrix carries the dimension of the industries-by-commodities. Neither the “Make” table nor the “Use” table is necessarily square; they are presented in different dimensions, which implies that the same commodity can be made and used in more than one industry.

The TFP from the NBER–CES database measures is not adjusted for capital utilization, as mentioned in Bartlesman and Gray (1996). An alternative approach is to obtain a measure of TFP that controls for aggregation effects, varying utilization of capital and labor, nonconstant returns and imperfect competition. See, for example, Basu et al. (2006) and Fernald (2014) for discussions.

Alternatively, one can estimate the VAR model in levels. The main purpose of this exercise is to construct instruments for endogenous variables in (2). Such an aim does not necessarily require us to take into account long-run relations between variables in the VAR model. We choose to estimate the VAR in growth because the idiosyncratic TFP measure in (2) is in the growth form. This choice could be problematic only if it results in weak instruments. However, as shown below, this is not the problem that we encounter.

See Ramey (2016) for a discussion of different identifications of technology shocks.

The impulse response functions to an unexpected positive TFP shock to each of 385 sectors can be found in the working paper version.

As mentioned above, in order to deal with the problem of endogeneity, Acemoglu et al. (2016a) suppress the contemporaneous effects of shocks and analyze only the lagged effects of shocks on the sectoral production.

Using the Wooldridge (1995)’s robust score test and a robust regression-based test also results in a strong rejection of the null hypothesis that \(Idio_{i,t}\), \(Down_{i,t}\), and \(Up_{i,t}\) are exogenous.

AAK consider a smaller sample 1991–2009. Our findings remain when using this sample, as shown in “Appendix 7.2.4”.

In future research, we would like to provide a rigorous rationale for this empirical feature.

F statistics are again substantially greater than the rule of thumb of 10, indicating that these instruments are relevant and do not seem to be weak.

Real sectoral output is calculated by dividing the nominal sectoral shipment value by the corresponding shipments deflator.

Excluding the lagged dependent variable does not alter our results.

Kose et al. (2003), Del Negro and Otrok (2008), and Mumtaz et al. (2011) also use this type of model to investigate the common dynamic properties of business-cycle fluctuations across countries. Meanwhile, Barauskaite and Nguyen (forthcoming) use this model to obtain the aggregate TFP shocks from sectoral TFP series.

Equivalently, one can permute rows and columns of input–output matrices.

The coefficients of time dummies and intercept are also centered around 0. Given that these coefficients are not of any interests, we do not show them.

Overidentified models are those that have the number of instruments exceeding the number of endogenous regressors.

The robust score test statistic is a version of Sargan’s test statistic for robust estimator of variance.