350 likes | 455 Views
Technology and the Business Cycle. Susanto Basu Boston College and NBER. EABCN Conference: Productivity and the Business Cycle. Helsinki, November 28, 2005. Draws on joint work with:. John Fernald Federal Reserve Bank of San Francisco Jonas Fisher Federal Reserve Bank of Chicago
E N D
Technology and the Business Cycle Susanto Basu Boston College and NBER EABCN Conference: Productivity and the Business Cycle Helsinki, November 28, 2005
Draws on joint work with: John Fernald Federal Reserve Bank of San Francisco Jonas Fisher Federal Reserve Bank of Chicago Miles Kimball University of Michigan
Technology and cycles: Early models • Early DSGE models uniformly assume technology shocks drive business cycles • Barro-King (1984): Hard to get consumption and leisure to move in opposite directions • Technology shocks can, for some parameter values • Plosser (1989) argues basic RBC model fits U.S. time series data well
Technology and cycles: Recent evidence SVAR literature: Blanchard-Quah (1989), Gali (1999), Francis-Ramey (2003) all find: • Technology shocks unimportant for fluctuations • Positive shock leads to a short-run decline in hours(Recent contrary evidence focus of Fernald’s talk) Direct measures of technical change:Basu-Fernald-Kimball (2004) [BFK], Shea (1998) find broadly the same results, especially the fall in hours
What Does Aggregate TFP (dp) Measure? • Technology change, dz • Unobserved changes in utilization du • Capital’s workweek, labor effort • Scale effects, • Reallocation of inputs across uses, R
Estimating Technology Change • Wanted: • Procedure to purge Solow residual of non-technological effects • Solution: • Estimate sectoral technology change using procedure robust to: • Increasing returns/imperfect competition • Variable utilization, • Non-observable wages at high frequency • Then add appropriately across sectors
BFK Method • Estimate industry IV regressions to get industry technology residuals dzi: • dyiis output growth • dxiis share-weighted average growth in observed inputs • dhiis growth in (detrended) hours per worker • Scale dzi to be on a value-added basis dziV • Aggregate technology, dz, is weighted sum of residuals:
On impact, output unchanged, hours fall With a lag, strong growth
Interpreting the evidence: I • Models with countercyclical markups can meet theBarro-King challenge • Lower markups raise real wages, so consumption and leisure may move in opposite directions • Sticky prices are one way to generate countercyclical markups • Gali and BFK interpret decline in hours after technology improvement as consistent with sticky prices • Suppose AD is given by Y = M/P, with M and P fixed • Technology improvements are contractionary if central bank does not accommodate sufficiently • Direct evidence from Marchetti-Nucci (2004)
Interpreting the evidence: II • Hours could easily go down due to strong income effects • Harder to reproduce “overshooting” found in data • Strong aggregate demand inertia (habit formation, investment adjustment costs) can explain the findings • Output cannot rise on impact • Therefore variable inputs fall temporarily • So can some “machine replacement” models—factories being retooled produce no output, and need little input • Behaviour of investment is much more informative than the response of hours for distinguishing among models
What are the sources of output growth? • Standard RBC models and standard empirics consider a single technology shock affecting both consumption and investment sectors • In practice, may be a variety of shocks (technology terms both secular and transitory) affecting different sectors, including the consumption and investment goods sectors • Does it matter? • Probably. Kimball (1994) • If the key to distinguishing models is the behaviour of output components, then we need to think through these issues
Simplest two-sector model with preferences allowing balanced growth Welfare thms say that the social planner’s problem above is also the solution to the competitive equilibrium
Previous Problem a Special Case of Following Social Planner’s Problem Equivalent problem:
Implications from the theory • Technology shocks to the consumption sector have no effect on N or I. (A can follow arbitrary stochastic process) • Intuition: Shocks to A change C, but since income and substitution effects cancel out, changes in C have no effect on labor supply or investment • However Z shocks will still have standard RBC effects
Further Results • With general King-Plosser-Rebelo preferences: If A follows a geometric random walk it has no effect on the optimal decision rules for N and IBFK cannot reject random walk for aggregate technology • The results extend to the case where there are different neoclassical production functions (different factor intensities) in the production of C and I
Why does it matter? • If we think that investment goods are produced with different technology (and we do) then we need to measure the “right” technology shocks to check investment predictions • When we measure technology in the data, either directly or through the long-run restrictions of VARs, what we capture is a combination of A and Z • So perhaps the reason why technology shocks appear not to matter for business cycles is that many are A shocks • Of course, in this case interpretation is not clear. Perhaps most technology shocks are A shocks, and then in response to “most technology shocks” we shouldn’t expect business cycle dynamics
A novel test of price stickiness • In the log case, a change in consumption technology should have no effect on investment and hours • For plausible deviations from log utility of consumption and permanent technology shocks (lower EIS and AR(1) technology), improvements in consumption technology should raise investment • But with sticky prices, Basu-Kimball (2001) show that improved consumption technology should lower investment and hours in the short run • Reason is that with price stickiness, relative price of consumption cannot jump down on impact • However, consumption technology should have RBC-style effect once price stickiness period passes
Measuring Sectoral Technology • Aggregate the industry shocks estimated by BFK into “consumption” and “investment” technology shocks • Use standard TFP for gov’t, mining, etc. • Need to use input-output tables to weight and aggregate • Same objective as Greenwood, Hercowitz, Krusell (2000) • Use quantity instead of price data • Can think of situations (e.g., markup shocks, factor market imperfections, different factor intensities) where relative price change doesn’t measure relative technology change • Theoretical findings implicit in GHK, but not easy to uncover given their normalization
Terms of Trade as Technology • In a closed economy, relative prices are always driven by domestic factors, including domestic technology • But this is not true with an open economy—the relative price faced by a small open economy can change due to changes in foreign technology or demand • We classify such price changes as “technology shocks” because they enable home consumers to have more consumption with unchanged labor input • View trade as a special (linear) technology, with terms of trade changes as technology shocks • However, this type of technology is special—for one thing, it has very different trend growth
Issues in using industry/commodity data to measure sectoral technical change • Final use is by commodity, productivity data are by industry • I-O make table maps commodity production to industries • Can translate industry technology into final-use technology, using dzCommodity = M-1dzIndustry, where dzIndustryis vector of industry technologies • Industry/commodity TFP is in terms of domestic production, whereas final-use reflects total commodity supply • Domestic commodity production plus commodity imports • I-O use table tells us both production and imports • Requires rescaling domestic-commodity technology
Data • Jorgenson et al (Fraumeni, Ho, Gollop, Stiroh, etc) KLEM productivity data as well as • Annual input-output (use and make) data underlying those productivity data • 1959-96 • Includes final use • Consumption is ND-S • Investment includes consumer durables
Data caveats • Underlying I-O and industry output data are from BLS, not from BEA (and don’t always match) • Final use data are not fully benchmarked to BEA • Doesn’t have full BEA hedonics • Let alone further Gordon/Cummins-Violante corrections
Prices of consumption relative to investment(percent change, annual rate) Note: Relative price of non-durable consumption to investment plus consumer durables. Software investment and housing services are excluded from BEA calculation for comparability with Jorgenson dataset.
Dynamics of growth in GDP with BFK residuals (preliminary) Heteroskedastic and auto-correlation robust standard errors in parentheses. Sample is 1961-1996. dz_bfk_ag is share-weighted sectoral technology shocks, using BFK residuals. Dz_bfk_ag differs from series actually used by BFK because it includes agriculture, mining, and government, which they omit; for those residuals, technology is taken to be standard TFP.
Dynamics of growth in non-residential investment with BFK residuals (preliminary) Heteroskedastic and auto-correlation robust standard errors in parentheses. Sample is 1961-1996. dz_bfk_ag is share-weighted sectoral technology shocks, using BFK residuals. Dz_bfk_ag differs from series actually used by BFK because it includes agriculture, mining, and government, which they omit; for those residuals, technology is taken to be standard TFP.
Dynamics of growth in non-dur consumption with BFK residuals (preliminary) Heteroskedastic and auto-correlation robust standard errors in parentheses. Sample is 1961-1996. dz_bfk_ag is share-weighted sectoral technology shocks, using BFK residuals. Dz_bfk_ag differs from series actually used by BFK because it includes agriculture, mining, and government, which they omit; for those residuals, technology is taken to be standard TFP.
Dynamics of growth in private business hours worked with BFK residuals (preliminary) Heteroskedastic and auto-correlation robust standard errors in parentheses. Sample is 1961-1996. dz_bfk_ag is share-weighted sectoral technology shocks, using BFK residuals. Dz_bfk_ag differs from series actually used by BFK because it includes agriculture, mining, and government, which they omit; for those residuals, technology is taken to be standard TFP.
Empirical “to do” list • Benchmark to BEA prices • Explore sensitivity to hedonics (e.g., Gordon) • Requires adjusting commodity output and capital inputs • Extend to more recent data • The productivity acceleration in the U.S.
Conclusions • Behaviour of output components following a technology shock is important for distinguishing among models • Theory predicts & evidence says that output components respond very differently to shocks to different sectors • E.g., consumption-technology neutrality • Going forward, need to construct theories that allow for differential sectoral technology shocks to see how we can explain our findings • Must reconsider empirical technology shock literature, SVAR-based as well as direct measurement methods • Need to keep at it—technology is almost the only “identified shock” that explains a large fraction of output variance at business-cycle frequencies