1. 1 Sector-Specific Technical Change Susanto Basu Boston College and NBER Jonas Fisher Federal Reserve Bank of Chicago John Fernald Federal Reserve Bank of San Francisco Miles Kimball University of Michigan and NBER

2. Motivation: What causes business cycles? Comovement: Candidate shocks must change Y, H, C, I in same direction Technology shocks can do this in simple models But in data, technology improvements often found to raise Y, C but reduce hours [Gali (1999), Basu, Fernald, Kimball (2006)] Francis and Ramey (2005): “The original technology-driven … business cycle hypothesis does appear to be dead.” Can we resuscitate technology shocks as major drivers of cycles? …while explaining the original negative evidence of Gali, BFK, et al? Yes…by reconsidering what an “aggregate” shock is Greenwood, Hercowitz, and Krusell and literature that followed (e.g., Fisher) emphasize final-use sector

3. We empirically measure technology by final-use sector: Consumption versus Equipment Investment A new empirical method to identify sector-specific shocks Idea: Estimate technology residuals from industry data, then aggregate through the input-output tables Contribution: We do not identify final-use technology from rel. price data Robust to time-varying markups, sticky prices, variable factor utilization, increasing returns, differing factor shares, changing tax rates, … Consumption- and investment-sector shocks individually have economic effects with right comovement properties Investment technology improvements are sharply contractionary, whereas consumption technology improvements are expansionary Prior empirical literature mostly considers the average of these shocks, which does not have the right comovement properties Relative prices respond very slowly to relative technology shocks Discuss what models might be consistent with the estimated effects

4. Outline 1.Motivation 2.Conceptual issues in empirical measurement 1.Mapping simple dynamic model to complicated world 2.Manipulating input-output (I-O) tables 3.Data and empirical results 4.Two stylized models 5.Conclusion

5. A general two-sector production structure Production functions for consumption, C, and investment, J: Production structure could be embedded in many DSGE models Want empirical estimates of Z C and Z J using assumptions consistent with large class of models

6. Note: Production structure is benchmark Greenwood- Hercowitz-Krusell, with different normalization of shocks We have separate technologies for the two sectors: Investment-specific technical change literature normalizes differently:

7. 7 7 Use relative price of cons. to equip. to infer relative technology of equipment to consumption How to measure Z C and Z J ? Existing method pioneered by Greenwood, Hercowitz, and Krusell

8. 88 Do relative final-goods prices reflect relative technologies? If identical production functions (especially factor shares)… …and identical factor price movements… …and identical time variation in markups and taxes … then relative price changes give relative technology change.

9. Our method: Estimate final-use technology “directly” without using relative prices Problem: 2-sector production model is simple, but data are complicated and messy Many firms/sectors, that sell to one another. Conceptually, consider stream of production leading to a final consumption/investment good, e.g., C i Function of the capital, labor, and technology used in entire stream But we don’t observe the K and L needed to produce each final output Use input-output tables to infer needed K, L and hence implied final- goods technology 9

10. Using input-output tables to map disaggregated technology shocks into final-use technology Direct technology estimates from industry production functions vector dz of (gross-output) technology shocks, [dz 1, dz 2, …]’ Implicit production function for delivering output to final consumption or investment. Intuition: Matrix B is (nominal) intermediate input shares b ij is share of commodity j in producing commodity i Technology for deliveries to final demand Weight by final-use shares, e.g.: 10

11. 11 Given TFP for final-use commodities, Z C, Z J, etc. easy

12. What is “net exports technology” In data, have to confront that economy is open Some commodity supply is imported Purpose of exports is to import (allowing use of those commodities) “Technology”: Terms of trade Final-use net-exports technology captures ability to obtain imports from exports Terms of trade improvements Technology improvements in goods we export 12

13. Outline 1.Motivation 2.Conceptual issues in empirical measurement 1.Mapping simple dynamic model to complicated world 2.Manipulating input-output (I-O) tables 3.Data and empirical results 4.Two stylized models 5.Conclusion

14. 14 Start with KLEM productivity data from Jorgenson et al. Key collaborators include Fraumeni, Ho, Stiroh, Gollop, and others Annual input-output tables underlying these productivity data 1960-2005 35 industries/commodities Includes final use, which allows us to distinguish ND-S Consumption (don’t have owner-occ housing) Consumer Durables Government purchases of G&S (not govt administration) Equipment investment Structures investment Exports and Imports

15. 15 We modify original data to incorporate alternative deflators for durable goods Key work of Gordon (1983), updated by Cummins-Violante (2002) New deflators redefine output for each industry Aggregate using I-O tables to get new measures of C, I, etc. Of course, also new prices for each category of expenditure

16. Need vector of industry technology innovations Production function Could use industry Solow residuals: Concerns: Non-constant returns unobserved variations in labor effort E i and capital’s workweek S i 16

17. 17 We estimate industry technology residuals dz i following Basu-Fernald-Kimball (2006, AER) Regress industry output growth dy i on input growth dx i and hours-per worker growth dh i : Use updated Ramey-Hall instruments: Hamilton-style oil-price increases, government defense spending, monetary innovations from an SVAR “Corrected” technology dz i = (c i +ε i ) controls for factor utilization and non-constant RTS For agriculture, mining, and govt enterpirises, where BFK don’t estimate residuals, we use uncorrected TFP Also use an unadjusted terms of trade

18. 18 Feeding industry BFK shocks through I-O tables: Equip and con. dur. technology rise fastest Cumulated log change in final-use BFK technology

19. Consumption and investment technology (exogenous component—zero out ToT, natural resources) 19

20. 20 Relative sectoral technology diverges from typical macro proxy of relative prices Relative price changes have correlation (in annual data) of only 0.23 with relative BFK technology Correlation of growth in relative TFP, relative BFK technology, and relative output prices Relative TFP Relative BFK Technology Relative Final- goods prices Relative TFP (dz Equipment – dz Consumption ) 1 Relative BFK Technology (dz BFK, Equipment – dz BFK, Consumption ) 0.281 Relative final-goods prices (dp Consumption – dp Equipment ) 0.760.231

21. 21 Does typical orthogonality assumption between “neutral” (consumption) and “investment-specific” technology hold? Correlations of final-use BFK Technology Subscripts: J is overall investment, JE is equipment and software, C is nondurables and services consumption. Sample is 1961-2005. Correlation Equipment investment (dz JE ) and consumption (dz C ) 0.64 “Investment specific” (dz JE -dz C ) and consumption (neutral, dz C ) 0.03 “Investment specific” (d z JE -dz C ) and equipment (dz JE ) 0.48

22. 22 Equipment investment technology and consumption technology have very different macroeconomic effects 22 Each row is a separate regression of log change in variable shown on current and lagged tech shocks Equip tech. includes con dur and govt equip. Cons. (Nondur) tech includes structures and nonequip. govt. Intrumental variables estimation. Instruments zero out terms of trade and industry shocks not estimated via BFK. Annual data 1961-2005.

23. 23 Technology shocks explain a lot of the variation in equipment… Corr = 0.59

24. 24 …as well as hours Corr = 0.64

25. 25 Alternative empirical approach: Fisher SVAR impulse responses look similar to BFFK pre-1980 25 Notes: Response of hours worked from a five-variable VAR (identified with LR restrictions). Variables are growth in relative equipment price; growth in labor productivity; hours per capita; consumption-price inflation; and 3-month T-bill rate. First sample is 1955:I-1979:II. Second sample is 1982:3 -2000:4.

26. 26 BFK (2006): Effects of aggregated technology shocks don’t look as much like business cycles 26 Annual data 1961-2004. Dependent VariableBFK Technology Change Lags012 GDP -0.170.190.07 (0.11)(0.12)(0.09) Investment (equipment) -0.660.060.43 (0.44)(0.42)(0.31) Consumer NDS -0.060.11**0.00 (0.08)(0.05)(0.07) Consumer Durables 0.070.50-0.08 (0.29)(0.30)(0.31) Hours -0.32**0.030.11 (0.14)(0.11)(0.08) Price of GDP -0.23**-0.15**-0.05 (0.11)(0.07)(0.06) Fed Funds Rate -0.41**-0.48**-0.27** (0.18)(0.11)(0.08) 10-yr Treasury yield -0.20-0.37**-0.32** (0.14)(0.12)(0.11)

27. 27 Why are investment technology improvements contractionary—even for investment? Relative price of I drops very slowly in response to relative technology improvement Creates strong incentive to delay purchases Durable goods have high intertemporal elasticities of substitution Demand for durables declines as a result

28. 28 In long run, relative prices (CNDS to equip/condurables) move with relative TFP and relative technology 28

29. 29 Relative prices respond to relative technology with long lags 29 Relative Price (LHS var): growth in price of consumption (ND and services) relative to price of equipment Rel. Technology (RHS var) : Growth in equipment technology relative to consumption (ND and services) d Relative Technology (Lag)Cumulative 0123effect d Relative Price 0.280.150.170.23 0.83 (cons to equip) (0.13)(0.20)(0.14)(0.19)(0.31)

30. 30 In Fisher’s SVAR, relative prices also respond very slowly to sectoral technology shocks 30

31. 31 Outline 1.Motivation 2.Conceptual issues in empirical measurement 1.Mapping simple dynamic model to complicated world 2.Manipulating input-output (I-O) tables 3.Data and empirical results 4.Two stylized models 5.Conclusion

32. 32 What models might match these facts? Our work uncovers two technology shocks that move output, hours, consumption and investment in the same direction To get output and other variables to increase, we need: a positive consumption technology shock or a negative equipment technology shock!

33. 33 In benchmark RBC model, consumption-specific technology shocks have no dynamic effects Suppose period utility is logarithmic U = ln(C) + v(1-L) Let A be multiplicative technology that affects only production of non- durable consumption Consumption-technology neutrality proposition: In two-sector RBC model, stochastic process for A does not affect labor hours L, investment J, or the quantity of resources devoted to producing consumption goods (X) A enters separably in the utility function, so it doesn’t affect decision rules

34. 34 Social-planner’s problem for two-sector growth model: A t is additively separable, and thus doesn’t affect decision rules

35. 35 Social-planner’s problem for two-sector growth model: A t is additively separable, and thus doesn’t affect decision rules Equivalent problem, where C=AX:

36. 36 Comments More general King-Plosser-Rebelo preferences: If A follows a geometric random walk it has no effect on optimal decision rules for L, X and J. Shocks to investment technology should have standard RBC effects We find neither consumption technology neutrality nor expansionary effects of investment technology improvements We may have resuscitated technology-driven business cycles, but not real business cycles!

37. 37 Simple 2-Sector Sticky-Price Model Period utility is: Dixit-Stiglitz preference for variety, implying markup of 1..1 Cobb-Douglas production functions for C and I Same fixed cost of production (10 percent of st. st. output) Same factor shares (  = 0.3) Factors mobile across sectors Calvo-pricing : Probability θ C = θ J =0.75 of keeping unchanged price Monetary policy follows Taylor rule, where Fed targets the “marginal cost gap” and consumer inflation

38. 38 Response to investment technology improvement labor

39. 39 Response to consumption technology improvement labor

40. 40 Conclusions Theory suggests that “final use” sector where technology change occurs matters for its effects E.g., consumption-technology neutrality, 2-sector sticky-price model Measure sectoral technical change using a new method that doesn’t require relative prices Effects of sector-specific technology shocks look like business cycles Equipment technology improvements reduce output, hours, investment, and consumption Consumption technology improvement raise output, hours, investment, and consumption Need to think about models that might explain these findings Multi-sector sticky-price models appear to have some promise

41. 41

42. 42 Empirical Implications: Low EIS and Permanent Tech Shocks With permanent technology shocks and King-Plosser-Rebelo utility and relatively low elasticity of intertemporal substitution (≈ 0.3), investment technology shocks also have very little immediate effects on labor hours, though they do raise investment in a way that consumption technology shocks do not.

43. 43

44. 44 Comments With log preferences, ln(A) is additively-separable: Any stochastic process for A has no effect on optimal decision rules for N, X and I. More general King-Plosser-Rebelo preferences: If A follows a geometric random walk it has no effect on optimal decision rules for N, X and I.

45. 45 What is “technology”? Is ‘technology’ the economy’s PPF? The change in production functions for domestic C and I? We use the first, broader, definition.

46. 46 Notes Trade technology is the terms of trade Suppose there are no intermediate-inputs and one of each final-use commodity (e.g., a single consumption good) Final-use technology is technology in that commodity Our definition is correct for typical two-sector macro model Otherwise, takes account of intermediate-input flows If all sectors face same input prices and have identical factor shares (including intermediates), then relative final-goods prices reflect relative technologies Again, our definition is correct for typical special cases used in macro (e.g., Greenwood, Hercowitz, Krusell)

47. 47 Two issues arise in using industry/commodity data to measure final-use sector technical change Final use is by commodity, productivity data are by industry I-O make table maps commodity production to industries We assume: MAKE*dz Commodity = dz Industry, Final-use is from total commodity supply, not domestic production I-O use table tells us both production and imports

48. 48 We aggregate commodity technology shocks to final uses with constant-share aggregation

49. 49 Motivation: In benchmark RBC model, consumption- technology shocks are neutral Suppose utility is logarithmic U = ln(C) – v(L) Let A be multiplicative technology for producing non-durable consumption Consumption-technology neutrality proposition: In two-sector RBC model, stochastic process for A does not affect labor hours L, investment J, or the quantity of resources devoted to producing consumption goods (X) A affects only production of nondurable consumption goods

50. 50 Social-planner’s problem for two-sector growth model, with CRS, identical production technologies

51. 51 This is special case of following problem, where A t is additively separable, and thus doesn’t affect decision rules

52. 52 This is special case of following problem, where A t is additively separable, and thus doesn’t affect decision rules Equivalent problem: Empirically, do shocks to different final sectors have different economic effects?

53. 53 Relative sector-specific TFP reflects relative output prices and relative primary input prices “Primal” (standard) TFP for commodity i equals dual: In vector/matrix form: Intermediate input prices are s N dn ≈ Bdp, so:

54. 54 What we do instead Seek a more robust way to measure relative technology Use industry data to estimate underlying shocks Production-function regressions a la BFK (2006) Then aggregate using I-O tables to final-use technology changes for C, I, etc. Present findings, implications for business-cycle models

55. 55 Outline 1.Introduction: Declining relative price of equipment 2.Motivation: Consumption-technology neutrality 3.Conceptual issues in empirical measurement 1.Mapping simple dynamic model to complicated world 2.Terms of trade as a form of technology 3.Manipulating input-output (I-O) tables 4.Data and empirical results: Bottom-up v. top-down 5.Interpretation

56. 56 Motivation: In benchmark RBC model, consumption-specific technology shocks have no dynamic effects Suppose period utility is logarithmic U = ln(C) + v(1-L) Let A be multiplicative technology that affects only production of non- durable consumption

57. 57 Note: This model is benchmark Greenwood-Hercowitz-Krusell model, with a different normalization of the two shocks We normalized on: Investment-specific technical change literature normalizes differently:

58. 58 We interpret terms of trade as a form of technology In closed economy, by definition, only domestic factors affect ability to convert consumption goods to investment goods In open economy, foreign technology or demand might affect ability to obtain consumption/investment with unchanged labor and investment Purpose of exports is to import trade is a special (linear) technology, with terms-of-trade changes as technology shocks

59. 59 Two issues arise in input-output data to measure relevant intermediate-input matrix B Final use is by commodity, productivity data (dz i ) are by industry I-O make table maps commodity production to industries Final-use is from total commodity supply, not domestic production I-O use table tells us both production and imports

60. 60 What does an input-output use table look like? Columns give inputs into domestic production Rows give “uses” of the commodity Nominal commodity-by-commodity use table

61. 61 We define a “trade goods” commodity, which uses commodity exports as an input to produce imports Nominal commodity-by-commodity use table Exports represent intermediate inputs into trade-goods production. Imports are used as intermediate inputs to produce commodity supply

62. 62 We define a “trade goods” commodity, which uses commodity exports as an input to produce imports Nominal commodity-by-commodity use table Exports represent intermediate inputs into trade-goods production. Imports are used as intermediate inputs to produce supplies of other commodities

63. 63 We define a “trade goods” commodity, which uses commodity exports as an input to produce imports Nominal commodity-by-commodity use table Exports represent intermediate inputs into trade-goods production. Imports are used as intermediate inputs to produce commodity supply

64. 64 Net exports are one use of trade goods, representing a claim on future imports Nominal commodity-by-commodity use table NX are a form of final expenditure, much like investment.

65. 65 Tables now add up, in terms of commodity supply! Nominal commodity-by-commodity use table

66. 66 We now transpose use matrix (for notational ease), and take row shares Use table, in (transposed) share form

67. 67 What is trade goods “technology”? The terms of trade We export in order to import Commodity exports are (intermediate) inputs into producing trade goods ‘Output’ is imports plus net exports nominal value = export value P X X Real output = Goods we can import = P X X/P M

68. 68 Sectoral TFP diverges from typical macro proxy of relative prices Relative TFP, output prices, input prices, and wedges (percent change, annual rate) Full Sample (1) 1960-1982 (2) 1982-2004 (3) Accel. ((3) – (2)) (1)Relative TFP (dz Equipment – dz Consumption ) 2.421.603.241.65 (2) Relative price (dp Consumption – dp Equipment ) 2.541.573.521.95 (3) Relative primary input prices (equip. rel. to NDS cons.) 0.190.090.290.20 (4) Relative import prices (equip. rel. to NDS cons.) -0.43-0.27-0.59-0.32 (5) Relative sales tax wedge (equip. rel. to NDS cons.) 0.110.210.02-0.19

69. 69 Does typical orthogonality assumption between “neutral” (consumption) and “investment-specific” technology hold? Correlations of final-use TFP Subscripts: J is overall investment, JE is equipment and software, C is nondurables and services consumption. 1960-20041960-19821982-2004 (1) Corr(dz J, dz C ) 0.830.900.75 (2) Corr(dz JE, dz C ) 0.740.820.67 (3) Corr(dz JE - dz C, dz C ) 0.180.27-0.02 (4) Corr(dz JE - dz C, dz J ) 0.610.580.54

70. 70 Equipment investment technology and consumption technology are quite positively correlated… Correlations of BFK “purified” final-use technology 1960-20041960-19821982-2004 (1) Corr(dz J, dz C ) 0.700.730.75 (2) Corr(dz JE, dz C ) 0.450.430.60 (3) Corr(dz JE - dz C, dz C ) -0.06-0.09-0.01 (4) Corr(dz JE - dz C, dz J ) 0.590.530.57 Subscripts: J is overall investment, JE is equipment and software, C is nondurables and services consumption.

71. 71 Equipment technology improves reduce output and hours— consumption technology improvements raise output Dep. Variables are growth rates of variables shown. Regressors are BFK final-sector technology shocks. Instruments are corresponding measures, with shocks to terms of trade, ag, mining, and govt. zeroed out. Std errors robust to heteroskedasticity and autocorrelation.

72. 72 Table 6

73. 73 Begin by using industry TFP Assume (for a start) that industry Solow (TFP) residual is the right measure of industry technical change Jorgenson data give us input-output (make) table B and the final-use vectors b

74. 74 E&S and con. durables TFP rises faster than for nondurables and services, government, or structures Cumulated log change in final-use TFP