Evidence from Firm-level Data in Vietnam

APPENDIX(1) Estimation of total factor productivity.The basic framework of the approach is as follows. The production function to be estimated is assumed tohave the form of Cobb- Douglas type, with labor and capital as input factors and value-added as output. Theestimation will be run across industries. The estimation equation of firm i at time t (in logarithmic form)isvt= β + β l + β k + ε(A.1)ol tkttwherevt, ltand k tare log of value-added, labor (the freely variable input) and capital (state variableinput) at time t , respectively, andεtis the error term whose explanation will come soon later. We drop thesubscript i for ease of expression. To produce, the firm uses also intermediate input, which is subtracted outof the total production to calculate the value-added. The predicted productivity is the exponent of the sum ofconstant coefficient and the error term. Stating differently, it is calculated as:TFPˆt= exp( v − ˆ β l − ˆ β k ) . (A.2)tl tktThe key point that is different from OLS estimation is that the error term has two components,ωtand ηt, where ωtis the transmitted productivity component that may be correlated with input choices andηtthe independently and identically distributed (iid) one that is uncorrelated with the choice of inputs.is not observable to econometricians, which leads to the problem of simultaneity stated in the main text.Those estimators that ignore this problem (like OLS) will have inconsistent results.ωtIn the framework of Levinsohn and Petrin (2003), the demand for intermediate inputm tis assumedto depend on the firm’s state variablektandωt:mt= m k , ω )(A.3)t(t tWith the assumption that this demand function is monotonically increasing inωt, we have ωtas a functionofktandmt:Now, the estimation equation can be rewritten asω = ω k , m )(A.4)tt(t tvt= β l + φ ( k , m ) + η(A.5)l ttttt25