The State and China s Productivity Deceleration: Firm-level Evidence

Transcription

1 The State and China s Productivity Deceleration: Firm-level Evidence Jorge Alvarez, Tuo Chen, Grace Li International Monetary Fund November 10, / 44

2 Motivation The Chinese SOE reform and privatization process slowed-down after the crisis The total factor productivity (TFP) of the manufacturing sector decelerated around the same time Empirical evidence shows that the privatization process accounted for a significant share of growth during the early 2000s (Hsieh and Song 2014 WP) A natural question: Can the TFP deceleration be explained by the reversal of the privatization/reform process? 2 / 44

3 Chinese privatization process and its reversal Figure 1: Shares of Capital by SOE category 3 / 44

4 TFP growth in China Figure 2: Firm-level Estimation and Aggregate Estimation 4 / 44

5 This paper Documents TFP dynamics (growth and deceleration) in Chinese manufacturing at both the aggregate and firm level Estimates TFP gaps between SOEs and private firms Assesses the role of SOEs in explaining aggregate TFP dynamics 5 / 44

6 Preview of the results The TFP growth trend in the manufacturing sector reversed in 2011 Within-firm TFP changes among SOEs were a major contributor to this reversal Improvements in resource allocation during the growth period across SOE firms seem to have stopped 6 / 44

7 Aggregate TFP 7 / 44

8 Data Aggregate Data: China Industry Statistical Yearbook Coverage: Contains value added, intermediate inputs, and labor Firm-level Data: Chinese Industrial Survey ( ) Coverage: Value added ( ), sales income, sales cost, and fixed assets Pseudo value added: sales income - sales cost 8 / 44

9 TFP measurement Aggregate level TFP estimation: Cobb-Douglas, constant returns to scale Firm level TFP estimation: Cobb-Douglas Olley-Pakes (1996): a ist = ω(k ist, inv ist,...) Levinsohn-Petrin (2003): a ist = ω(k ist, m ist,...) De-Locker (2011): correct for the potential price bias All TFP measures give the same trend and have a high correlation 9 / 44

10 Firm-level estimation: Olley-Pakes (1996) First Step: y ist = α + β l l ist + φ(k ist, inv ist ) + ε ist where φ(k ist, inv ist ) = β k k ist + ω(k ist, inv ist ) φ ist = y ist ˆβ l l ist ˆα Second Step: Assume a ist follows an Markov process: a ist+1 = g(a ist ) + η ist φ ist+1 = β 0 + β k k ist+1 + g(ω(k ist, inv ist )) + ν ist Use higher order polynomials to approximate the unknown function g(.) and ω(.,.) y ist is the log value of real value added, l ist is the log value of labor, k ist is the log value of real fixed asset Levinsohn-Petrin 2003 replaces inv ist by m ist 10 / 44

11 Firm-level estimation: De Loecker (2011) Monopolistic competition First Step: y ist = α + β l l ist + β s y st + φ(k ist, m ist ) + ε ist where φ(k ist, m ist ) = β k k ist + ω(k ist, m ist ) elasticity of substitution: ε s = 1 β s β l = β l 1 1+ β s β k = β k 1 1+ β s φ ist = y ist ˆβ l l ist ˆβ s y st ˆα Second Step: Assume a ist follows an Markov process: a ist+1 = g(a ist ) + η ist φ ist+1 = β 0 + β k k ist+1 + g(ω(k ist, m ist )) + ν ist Use higher order polynomials to approximate the unknown function g(.) and ω(.,.) y ist is the log value of real value added, l ist is the log value of labor, k ist is the log value of real fixed asset 11 / 44

12 Correlation of Different Measures Table 1: DlnTFP Correlation measured by VA Variables DL LP OP CD DL LP OP CD DL: De Loecker; LP:Levinsohn-Petrin; OP: Olley-Pakes; CD: Cobb-Douglas 12 / 44

13 Complications in firm-level data Value-added data is only available from 1998 to Pseudo-VA = Sales Income - Sales Cost 0.87 correlation with VA available , Table 2: DlnTFP Correlation measured by Pseudo VA Variables DL LP OP CD DL LP OP CD DL: De Loecker; LP:Levinsohn-Petrin; OP: Olley-Pakes; CD: Cobb-Douglas 13 / 44

14 Evolution of average TFP 1 (a) Directly Observed VA (b) Pseudo VA = Sales Income - Sales Cost Figure 3: Unweighted Mean of lntfp by different measures 14 / 44

15 Evolution of average TFP 2 (a) Directly Observed VA (b) Pseudo VA = Sales Income - Sales Cost Figure 4: Mean of lntfp by different weights 15 / 44

16 Evolution of average TFP 3 Figure 5: Firm-level Estimation and Aggregate Estimation 16 / 44

17 Not driven by sectoral composition year Mining Medicine Telecom Energy Transport Figure 6: Decomposition by Sectors 17 / 44

18 SOEs vs Private Firms 18 / 44

19 SOEs vs private firms: private firms are more productive Table 3: Sector Premiums log(tfp) SOE *** *** *** *** [ ] [ ] [-7.855] [-7.172] Fixed Effects Year Yes Yes Sector x Year Yes Yes Firm Yes Yes Number of obs 242, , , ,332 R / 44

20 SOEs vs private firms (cont.): controlling for size Table 4: Sector Premiums log(tfp) SOE *** *** *** *** [ ] [ ] [-8.213] [-8.057] Size *** *** *** *** [ ] [ ] [ ] [ ] Fixed Effects Year Yes Yes Sector x Year Yes Yes Firm Yes Yes Number of obs 242, , , ,119 R / 44

21 SOEs vs private firms (cont.) 1998 Density log(tfp) kernel = epanechnikov, bandwidth = SOEs Non-SOEs Figure 7: Firm-level TFP for SOEs and private firms 21 / 44

22 SOEs vs private firms (cont.) 2011 Density log(tfp) kernel = epanechnikov, bandwidth = SOEs Non-SOEs Figure 8: Firm-level TFP for SOEs and private firms 22 / 44

23 SOEs vs private firms (cont.) 2013 Density log(tfp) kernel = epanechnikov, bandwidth = SOEs Non-SOEs Figure 9: Firm-level TFP for SOEs and private firms 23 / 44

24 Evolution of SOE premiums year Figure 10: SOE Premiums 24 / 44

25 Evolution of SOE premiums year SOE Premium +sector,size +sector Figure 11: SOE Premiums 25 / 44

26 Decomposition 26 / 44

27 The role of SOEs in aggregate TFP What is the contribution of SOEs to TFP Dynamics? Three channels: Privatization: Movement of labor and capital away from SOEs Reallocation: Movement labor and capital across SOE firms Within-firm dynamics: TFP changes within SOE firms 27 / 44

28 TFP Decomposition TFP t A t = = Σ s Σ i s k α it lβ it K α t Lβ t Yt K α L β = Σ i A it k α it lβ it K α t Lβ t kit α Σ l β it i s Σ i s kit αl β A it it }{{} TFPt s = Σ s Σ i s k α it lβ it K α t Lβ t TFP s t 28 / 44

29 Decomposition of TFP by SOE/Non-SOE Category TFPt Σ i s kir α Σ l β ir kir α s Kr α L β Σ l β ir i s r Σ i s kir αl β A ir ir }{{} TFP r Σ i s kir α = Σ l β ir s Kr α L β r [TFP S r TFP S t ] } {{ } Within SOE/non-SOE changes Σ i s kit α Σ l β it s Kt α L β t Σ s (TFPr S TFPt S )( Σ i skir αl β ir Kr α L β Σ i skit αl β it r Kt α L β ) t }{{} Covariance term Σ i s kit αl β it Σ i s kit αl β it A it } {{ } TFP t + Σ s [ Σ i skir αl β ir Kr α L β Σ i skit αl β it r Kt α L β ]TFPr S t }{{} Between SOE/non-SOE changes 29 / 44

30 SOE vs Non-SOE Decomposition Individual TFP change decomposition Total Within Non-SOE TFP change Within SOE change Privatization Figure 12: TFP Changes by SOE Category 30 / 44

31 Between-Within firm decomposition TFPt S TFPr s TFPt s k = W stay s ir α,r Σ l β ir i stay s Σ i stay s kir αl β [A ir A it ] ir }{{} Within Firm k +W stay s ir α,r Σ i stay s [ l β ir W stay s,t W stay s,r kit αl β it Σ i stay s kit αl β ]A it } Σ i stay s kir αl β ir {{ it } Between Firm k +W enter s ir α,r Σ l β ir i enter s Σ i enter s kir αl β ir A ir } {{ } Entry kir α Σ i stay s (A ir A it )( l β ir Σ i s kir αl β k W exit s it α,tσ l β it i exit s Σ i exit s kit αl β it A it } {{ } Exit kα it l β it ir Σ i s kit αl β ) it }{{} Covariance 31 / 44

32 Between-within firm decomposition by SOE category TFP change decomposition TFP change decomposition Total Reallocation between firms Covariance Within firm TFPchange Exit/Entry Total Reallocation between firms Covariance Within firm TFPchange Exit/Entry (a) Within non-soe TFP Changes (b) Within SOE TFP Changes Figure 13: TFP Changes within SOE/Non-SOE 32 / 44

33 Firm-level decompositions: National SOEs and Local SOEs Table 5: Zoom into SOEs Component % % Within National SOEs % % - Within firm changes % % - Between firm changes % % - Exit / Entry % % - Covariance term % % Within Local SOEs % % - Within firm changes % % - Between firm changes % % - Exit / Entry % % - Covariance term % % 33 / 44

34 Conclusion There has been a marked deceleration in both aggregate and firm-level measures of manufacturing TFP in China Within-firm TFP changes among SOEs and privatization were drivers of aggregate growth from 1998 to 2007 This trend reversed after 2011 (or earlier) One narrative: after the financial crisis, SOE reforms slowed down 4 Trillion Yuan stimulus was directed to sub-optimal SOE investments 34 / 44

35 Next steps Explore credit and interest expense data to study the channels through which capital was reallocated across SOE firms Study the link between TFP deceleration and contemporaneous state financing programs such as the 4 Trillion Yuan stimulus package Study firms that transitioned out of SOE status using the panel dimension of the data Look into expanding data coverage 35 / 44

36 Cost of debt for SOEs Interest Rate year SOEs Non-SOEs Figure 14: Mean interest rate (interest expense / debt) 36 / 44

37 Next steps Explore credit and interest expense data to study the channels through which capital was reallocated across SOE firms Study the link between TFP deceleration and contemporaneous state financing programs such as the 4 Trillion Yuan stimulus package Study firms that transitioned out of SOE status using the panel dimension of the data Look into expanding data coverage 37 / 44

38 Misallocation in SOEs? Var(MPK) year SOEs Non-SOEs Figure 15: Variance of marginal product of capital 38 / 44

39 Next steps Explore credit and interest expense data to study the channels through which capital was reallocated across SOE firms Study the link between TFP deceleration and contemporaneous state financing programs such as the 4 Trillion Yuan stimulus package Study firms that transitioned out of SOE status using the panel dimension of the data Look into expanding data coverage 39 / 44

40 Appendix 40 / 44

41 Firm Decomposition: SOEs vs private firms Table 6: Sector Premiums - Unweighted Component % % Within SOEs % % - Within firm changes % % - Between firm changes % % - Exit / Entry % % - Covariance term % % Within Private Firms % % - Within firm changes % % - Between firm changes % % - Exit / Entry % % - Covariance term % % 41 / 44

42 SOEs vs private firms (cont.): weighted Table 7: Sector Premiums - Weighted log(tfp) Weighted SOE *** *** *** *** [ ] [ ] [7.518] [-6.187] Fixed Effects Year Yes Yes Sector x Year Yes Yes Firm Yes Yes Number of obs 242, , , ,332 R / 44

43 SOEs vs private firms (cont.): controlling for size and weighted Table 8: Sector Premiums - Weighted log(tfp) Weighted SOE *** *** *** *** [ ] [ ] [6.095] [-4.311] Size *** *** *** *** [ ] [ ] [ ] [ ] Fixed Effects Year Yes Yes Sector x Year Yes Yes Firm Yes Yes Number of obs 242, , , ,119 R / 44

44 Evolution of SOE premiums year SOE Premium +sector,size +sector (weighted) +sector SOE Premium (weighted) +sector,size (weighted) Figure 16: SOE Premiums 44 / 44