Erdem and Tybout (2003) deal with these limitations, albeit using artificial data. Their analysis is

Adrienne Claire

Informations

• Publié par: Zuim
• Nombre de lectures: 27
• Langue: English

Extrait

J. Tybout

   Economic Development 570B Lecture 25 December 4, 2003   A number of studies have found that plant sizes tend to shrink in the face of heightened foreign competition, just as the Melitz model predicts.   Nonetheless, the effects on scale efficiency are quite modest. The reason is that most of the adjustment in total domestic output comes from large firms that are operating in the flat portion of their average cost curves.   exp: Mexico Using plant-level panel data and econometrically estimated cost or production functions, It is possible to decompose industry-wide growth in average cost or total factor productivity into the effects of better scale economy exploitation, reallocation of market shares toward more efficient firms, and a residual, which reflects intra-plant improvements in efficiency. (In the case of cost functions the residual also reflects changes in relative prices.)   (graphs of scale economies here)   Decomposition of Average Cost and Productivity Growth in Mexico, 1984  1990  

Average Cost Growth (Percentages)    Scale Effect Share Effect Residual Effect Total Growth -0.79 0.98 -5.07 -6.84 - Productivity Growth (percentages)   Scale Effect Share Effect Residual Effect Total Growth 0.55 1.02 9.60 11.17

  source: Tybout, James and M. Daniel Westbrook. "Trade Liberalization and Dimensions of Efficiency Change in Mexican Manufacturing Industries," Journal of International Economics , August 1995, 39(1/2), pp. 53-78.   Note: Real output from the manufacturing sector expanded 54 percent during 1984-1990.   Let output at the i th firm in year t be given by q it = A it h ( v it ) , but now write h ( v it ) = ( g ( v it ) ) where g ( v it ) is a constant-returns homothetic function of the input n t vector, v it , and (⋅) captures any scale economies. Also, let S it = g ( v it ) / ∑ g ( v jt ) be this i = 1 firms market share in terms of its input use and let B it = q it / g ( v it ) be its productivity

n t level. Then the rate of growth in industry-wide average productivity, B t = ∑ B it S it , can i = 1 be decomposed as:   dB t = n t   dg it   −   q it   + n t   it   + n t d it q it B t i ∑ = 1  g it  ( µ it 1)  q t  i ∑ = 1 dS it  B B t  i ∑ = 1    A A it       q t           where it = d ln ( q it ) / d ln( g it ) measures returns to scale at the i th plant in year t .   Using cross-industry variation, one can try and link to foreign competition : measure with changes in ERPs, QRs, import penetration rates (M), or export rates (X). But its not a simple matter of identifying protected import-competing sectors . X and M are highly correlated in the base year (1984), implying lots of intra-industry trade. The sectors with high X and M also have relatively low license coverage and tariffs.   Movements toward more openness werent systematically related to initial level of openness. The open sectors had the biggest productivity gains  this was mainly a residual effect, but some share reallocation too. Scale effects were significantly correlated with DX and DM, but they go the wrong way  more trade leads to less scale efficiency. (Releasing constraint on imported inputs?)   Other studies have found that falling productivity dispersion accompanies liberalization.    exp: Chilean liberalization             AC                                               IS + IS Q                                                                                                                                                                             XO + XO Q                                                                                        Q                                                   Fitting the two regression lines, one from each economic census, get an intercept, slope, and measure of dispersion for each industry, before and after trade reforms. (Complications: measurement error, selectively missing capital stock data.)   If competitive pressures eliminated inefficient plants or forced them to improve, wed expect to find the intercept falling, the amount of dispersion falling, and the slope rising (becoming less negative).  

This didnt happen on average, but it did happen in the industries that underwent the largest reductions in effective protection   source: Tybout, James, Jaime de Melo, and Vittorio Corbo "The Effects of Trade Policy on Scale and Technical Efficiency: New Evidence from Chile," Journal of International Economics 31 (November 1991), 231-250.   Intra-firm productivity growth and reallocation effects   Chile liberalized during 1974-1979, then experienced overvaluation during 1979-82. The exchange rate regime collapsed in 1982, and the country went into severe recession. Thereafter Chile rebounded.   Nina Pavcnik (2002) asks: is there evidence that the heightened competition from abroad rationalized industries in the sense that Melitz describes?   Start with a standard Cobb-Douglas production function in unskilled labor, skilled labor, materials, and capital:   ln( Y it ) = β 0 + β 1 ln( L it ) + β 2 ln( H it ) + β 3 ln( M it ) β 4 ln( K it ) + ln( A it ) + u it       Here A it is the component of productivity growth that the entrepreneur anticipates one period in advance, and u it is the surprise component. The econometrician observes neither, so the compound disturbance is correlated with the factor input stocks, inducing simultaneity bias. Good instruments arent available, so Pavcnik deals with this problem by using the Olley-Pakes estimator. This amounts to treating capital investments, given the current capital stock, as a monotonic function of the productivity shock. Inverting this function, one can combine the last two terms into a non-parametric function of K it and obtain consistent estimates of the other coefficients. Then, with some extra details to deal with entry and exit (which causes selection bias), the capital coefficient can be retrieved.   Once the parameters of this function are obtained, industry by industry, one can construct a  relative productivity measure for each plant and year: µ it = ln A it + u  it − A 0 + u 0 . Here zero subscripts and overbars refer to the average base year values for the same industry.) Then, the weighted average productivity measure for the industry is:   W t = ∑ s it µ it = µ t + ∑ ( s it − s t )( µ t − µ t ) i i   That is, productivity growth is due to within (covariance) and between forces. If firms with above-average productivity gain market share, the latter is important. By this measure, aggregate productivity grew 19 percent over the period 1979-86, and 13 percent of this was due to growth in the within effect.   Furthermore, the most rapid covariance growth was in the import-competing sectors (21.3 percent), followed by exportables (16.6 percent) and the nontraded goods sectors (2.4 percent).

  Also, as expected, exiting plants are much less productive than continuing plants.   Other studies use less elaborate methodologies but find similar results  both intra-plant gains and reallocation-based gains are significant. (See Erdem and Tybout 2003 and Tybout 2003 for surveys of the literature.)   Finally, one can try to link adjustments to trade orientation using the following regression:   it = 0 + 1 time t + 2 trade it + 3 time t ⋅ trade it + 4 Z it + it        where time is a vector year dummies, trade is a vector of trade orientation dummies, and Z is a vector of plant characteristics such as industry affiliation and whether a plant ceases to produce in a given year.   Pavcnik finds that the import-competing sectors show the most improvement, and that they began from a relatively low productivity level. This conclusion is typical of this type of studies of trade liberalization episodes (see Erdem and Tybout, 2003, for a survey).   Should we conclude that trade liberalization is a good thing?   Caveats   • Without a structural model linking performance to trade, it is difficult to say what the causal factors are behind entry/exit and market share reallocations. One alternative interpretation is that the tradeable goods producers were larger and more heavily indebted in dollars. The government bailed out these producers with a special exchange rate for repayment. Or, it could be that real exchange rate fluctuations are the key here. Note that nontradeables are doing relatively well until the regime collapse in 1982, and the opposite is true for exporters.   • This exercise and others like it dont reveal the underlying mechanisms that induce efficiency changes. Is it: elasticity-based changes in the return to innovation a la Aghion et al, or changes in the nature of an agency problem (e.g., risk of bankruptcy and loss of rents), heightened international technology diffusion, or Grossman-Helpman type mechanisms? For policy, its very helpful to understand the transition mechanism and whether it is inherently trade-related.   • The productivity measure is real revenue per unit input bundle. This is OK if the assumptions of Eaton and Kortum are satisfied, but if not, this measure may not have much to do with productivity:   • We dont get a complete accounting of the gains from trade by looking at contemporaneous changes in the mix of firms. One should also figure in: losses in varieties if domestic firms exit, capital losses imposed on surviving firms, and employment turnover generated in the transition.  

Transition dynamics and welfare Erdem and Tybout (2003)   deal with these limitations, albeit using artificial data. Their analysis is based on the simulation framework developed by Pakes and McGuire (1994, 2001).   Ericson and Pakes (1995), Pakes and McGuire (1994, 2001)hereafter, PEMprovide a framework for simulations.   Features of PEM: • Partial equilibrium: differentiated product industry.   • Entrepreneurs create new firms when the expected discounted net earnings stream exceeds entry costs;   • Entrepreneurs sell their firms for scrap when the expected discounted net earnings stream is less than scrap value.     • Active firms Bertrand compete in the product market each period (logit demand functions). Price choices dont affect future states.   • Active firms can invest in R&D each period. The larger the investment, the higher the probability of a product improvement. (Strategies are Markov-perfect Nash.)   • The outside good improves with constant exogenous probability.   • The quality of entrants goods is drawn from a distribution that improves at the rate of improvement in outside goods.     Details Let the j t h element of the vector s indicate the number of active firms at quality level j . An active firm with product quality level i earns current period profits π ( i t , s t ) , and its value function satisfies:       V ( , ) = max φ , π ( , ) sup 0  −+β ′ , ′ ( ′ , ′ | , , ) i s      i s + x ≥   cx i ′ , ∑ s ′ V ( i s ) p i s x i s           p(i ,s | i,s) perceived transition probability distribution   firms scrap value x  firms investment level c  unit cost of investment.      New firms pay a stochastic sunk start-up cost, x e , and begin with initial quality i e  relative to the imported good.   They enter when their expected value V ( i e   ,s ) is positive. (New firms earn zero profits during their first period of operation.)    

Firms spend x t  to achieve a unit increment to their quality (relative to the imported good) with t ax t p .   robabili y ax t + 1      Equilibrium obtains when beliefs are consistent with actual behavior.   Our modifications to the PEM framework: Replace the outside good with an imported variety with exogenous price, P 0 , to compete with the domestic varieties. Use nested logit demand system.   The quality of the imported variety improves each period with probability, δ .   The distribution of initial product qualities improves with the quality of imported goods (embodied tech. change).   The j th consumer derives the following utility from consuming a unit of product i in period t :     ω 0 t − +ω it − + θ P it + + ξ j ,1 t − + (1 −ϕ ) ε ijt i = 1, N t U = ijt ω 0 t θ P 0 t ξ j ,0 t (1 ϕ ) ε 0 jt i = 0     Here it = f ( i t ) is excess mean utility derived from goods of quality i , beyond that obtained t ω 0 t g  ζ k from the imported good. f ' ( ⋅ ) ≥ 0 , f ' ' ( ⋅ ) < 0 . Note that after t periods, =   k ∑ = 1     is mean utility derived from the imported good.   The model allows for: Schumpeterian effects Escape competition effects Market share and entry/exit (rationalization) effects Embodied technological change It does not allow for agency/shirking effects   The policy experiments    Reduced Price of Imports, ↓   P 0 (RPM). Commercial policy reforms One-time appreciation Accelerated rate of innovation for imported goods, ↑   δ  (AIM)   Allow a different class of goods into the country; Trade with a different kind of partner.   Simulation details For both experiments, simulate 100 trajectories of 5,000 periods each and average.  

Start from negligible import penetration.   Typically, 25 percent of the firms turn over in one period, so think of one period as 2-3 years.   All firms have same marginal cost, but think of improvements in quality as similar to idiosyncratic reductions in marginal cost. (Both increase profits.)   Reduced Price for Imports Accelerated Innovation for (RPM) Imports (AIM) 1 1 10 10 0.925 0.925 0.1 0.1 21 21 2 2 5 5 0.5 0.5 1.5 to 0 1.5 0.6 0.6 to 0.8

Parameters Marginal costs of production (domestic firms) Market Size (M) Discount factor ( ) Scrap Value ( ) Max Efficiency ( i max ) Investment efficiency (a) Price sensitivity of consumers ( ) Degree of substitution between nests ( ) Price of the imported good ( P 0 ) Probability of Innovation in the Imported good )     How important are escape competition effects?   ( i + 1, s − i | P 0 = 0) − ( i , s − i | P 0 = 0) versus ( i + 1, s − i | P 0 = 1.5) − ( i , s − i | P 0 = 1.5)      ( i + 1, s − i | t = 1) − ( i , s − i | t = 1) versus ( i + 1, s − i | t = 0) − ( i , s − i | t = 0)   Calculate differences for 100 most common states in base case, 4 best firms.

 

                                       

Increment to Profits from Successful Innovation: After versus Before Import Price Decline

0.050 0.000 -0.050 0 -0.100 -0.150 -0.200 -0.250 -0.300 -0.350

1

2

3

4

Increment to Profits From a Unit Innovation: After versus Before Improvement in Imported Good Quality

1.200 1.000 0.800 0.600 0.400 0.200 0.000 -0.20 0 .000

1.000

2.000

3.000

4.000

5

5.000

  

       

Table 3: Summary Statistics for RPM and AIM regimes*

Reduced Price for Accelerated Import Base case Imports (RPM)   Innovation (AIM)   P 0 =1.5, P 0 = 0.0 , P 0 =1.5 = 0.6 0.6 0.8 55.8 57.8 72.1 3.8 3.6 2.9 5.3 5.0 4.2     855.3 1045.9 686.5*** 27.9 25.0 30.1 882.8 1072.0 716.6***

    Percentage of periods with entry and exit* Mean number of firms active* Mean lifespan*   Mean consumer surplus** Mean producer surplus** Mean total surplus**   * Means taken across 100 trajectories of 5,000 periods each   ** Means taken across 100 trajectories of 100 periods each, discounted back to initial year of regime ***Excludes gains due to more rapid growth in the average quality of goods.  

 

Figure 1.1: Figure 1.2: Dome stic M arke t Share , RPM Expe rime nt Unweighted Mean Efficiency, RPM Experiment

1.05 1 0.95 0.9 0.85 0.8 0.75

time Figure 1.3: Weighted Mean Efficiency, RPM Experiment 7.8 7.6 7.4 7.2 7 6.8 6.6 6.4 6.2 6

  10

time

4.9 4.8 4.7 4.6 rho=0 after t=50  4.5 rho=1.5 4.4 4.3 4.2 4.1 4

rho=0 after t=50 rho=1.5

time   Figure 1.4: Unweighted Mean Investment, RPM Experiment 0.9 0.8 0.7 0.6 rho=0 after t=50  0.5 rho=1.5 0.4 0.3 0.2 0.1 0

 

time

rho=0 after t=50 rho=1.5

  

 

2.5 2 1.5 1 0.5 0

   

Figure 1.5: Price -Cost M argin, RPM Expe rime nt

1.32 1.3 1.28 1.26 1.24 1.22

time

Figure 1.7: Producer Surplus, RPM Experiment

rho=0 after t=50 rho=1.5

rho=0 after t=50 rho=1.5

Figure 1.6: One Firm Concentration, RPM Experiment

120 100 80 60 40 20 0

0.73 0.72 0.71 0.7 0.69 0.68 0.67 0.66 0.65 0.64

time

rho=0 after t=50 rho=1.5

Figure 1.8: Consumer Surplus, RPM Experiment

rh=0 after t=50 rho=1.5

 

 

11