Table 2: Growth Regression, OLS

Transcription

1 Post-socialist transition Economics 1670-W Professor Berkowitz Fall Growth Accounting in Russia s Regions Notes for the First Assignment The paper by Berkowitz and DeJong (2003) entitled, Policy Reform and Growth in post- Soviet Russia develops a statistical model of growth in the average Russian region. The idea is to develop a model in which small enterprise formation is a determinant of growth as well as policy and initial conditions. Because, in fact, small enterprises might cause growth, or growth might attract small enterprise formation, we add an additional model of just how policy and initial conditions determine small enterprise formation. Here is the basic model. An explanatory variable with p-value of 0.10 or less means typically means that its impact is significant (i.e., likely to be non-zero). Here, we employ a p-value of 0.20 because we have many explanatory variables and only 48 observations. Table 2: Growth Regression, OLS Explanatory Variable Coefficient Estimate Error t statistic p value Constant Distance (log) Initial Income IO Defense Education Reformist Voting Price Liberalization Large-Scale Privatization Small-Scale Privatization New-Enterprise Formation R 2 : Note: errors are heteroskedasticity consistent (White, 1980).

2 The results in table 2 tell us that that variables that DIRECTLY matter for growth are IO, DEFENSE, PRICE LIBERALIZATION and NEW ENTERPRISE FORMATION. The rest of the variables may matter INDIRECTLY via their impact on NEW ENTERPRISE FORMATION. These potentially indirect determinants of growth are Initial Income, Reformist Voting, Education, Small Privatization and Large Privatization and Log Distance to Moscow. This model then is estimated in Table 3. Panel A is NEWENT = Constant + a1*log-distance + a2*initial Income + a3*education + (1) a4*ref-voting + a5*large Priv + a6*small Priv GROWTH = Constant + b1*io + b2*defense + b3*price Lib + (2) b4*newent(pred) Where NEWENT(pred) is the level of new enterprise formation predicted in equation (1). Note that Constant, a1-a6 and Constant, b1-b4 are all estimated. Variables that considered statistically significantly different than ZERO have p-values typically less than For reasons that will be explained in class, we believe that b3 (the price liberalization coefficient) is statistically significantly different than ZERO even though it has a p-value of You can compute that quantitative significance of each variable by multiplying its estimated coefficient times it sample standard deviation (reported at the bottom of Appendix Table A1). For example, the quantitative significance of (a one-standard deviation increase in) Education on New-Enterprise Formation is 0.319*4.2 = You can also compute the indirect impact of (a one -standard deviation increase in) Education on Growth. This generates new enterprises. Checking the growth regression, each additional enterprises generate 1.04 percentage points growth per annum. Thus, the indirect impact of EDUCATION on growth is 1.354*1.04 which is about 1.36! The statistical model of growth works on average! That is, plug in the sample average values of IO, Defense, Price Lib and NEWENT into the growth regression and the

3 PREDICTED VALUE will equal the Sample average (-1.79). The same happens with the NEWENT equation. For example, with NEWENTS, plugging the average values, then NEWENT(on average) = (-0.135)*ln(1926) + (0.695*2.71) * * * (-0.231*0.331) = 4.3 (which is very close to the average of 4.36, and is only a bit off because of rounding off!!!... note, that I use the natural log and not the base 10!!!) However, if you compute the predicted NEWENT and GROWTH outcomes for your particular region, then the model may over or under predict. For example, consider VOLOGRAD. Compute the level of NEWENT predicted by the model: NEWENT(predicted) = (-0.135)*ln(1068) + (0.695*2.32) * * * (-0.231*0.295) = But, check the descriptive stats in Table 1, and you will find that the ACTUAL VALUE for Volgograd is So, the model under-predicts new enterprise formation in Volgograd In other words, given its distance from Moscow, Initial Income, Education Level, Reformist orientation and Privatization record, we would expect to see MANY FEWER(!!!!) small enterprise registries in Vologograd at the beginning of Just how bad is this under-prediction. Again, check table 1, and you will that the sample standard deviation for small enterprise formation is The extent of underprediction is = 1.96 which is less than a sample standard deviation. So, in fact, the model is pretty good (i.e. less than a sample standard deviation off) in this case. You can do the same for growth. The key point is that here we PLUG in for the PREDICTED VALUE OF NEWENT (4.2) instead of the actual value (6.14). This is how we solve the problem (as Molly and Brian noted in class) of determining the EXTENT to which NEWENT influences GROWTH as PREDICTED BY THE BERKOWITZ-DEJONG MODEL! The model predicts that growth in Volgograd is GROWTH(predicted) = * * * *6.14 = -2.99%

4 The actual growth record in Volgograd is 2.66%. So, even after using the predicted level of NEWENT, we are under-predicting growth in Volgograd by 2.66-(-2.99) = 5.67%! That is substantial amount in an economy where most regions are shrinking! More precisely, one standard deviation of growth is 4.44%, so, if we are sticklers, this does not seem too bad (bad would be more than 2 standard deviations). But, in reality, this does quite poorly, and this says that while our model does well on average, it is missing something quite fundamental in Volgograd. So, there is much to learn by getting sources on how reform progressed in Volgograd. You want to perform this statistical baseline for your own region. Think just how do its initial conditions and policy choices compare with the other regions in Russia? What dos this model say or miss about growth in your region? And, after doing this baseline and checking your other sources, just what factors can explain economic performance in your region? A model that includes foreign investment in the model is in Accounting for Growth in post-soviet Russia, Berkowitz and DeJong (2002). The papers Policy Reform and Growth in post-soviet Russia and Accounting for Growth in post-soviet Russia, cover 48 of Russia s 89 regions and measure growth between 1993:IV and 1997:IV. Both of these papers are available at Coverage for 70 regions that explain growth between 1993:IV and 2000:IV is in Enterpreneurship and Post-Socialist Growth also by Berkowitz and DeJong (2004). This paper is also available at

5 Table 3: Two-Stage Least-Squares Estimates New-Enterprise Regression Explanatory Variable Coefficient Estimate Error t statistic p value Deviation of Variable Quantitative Significance Constant Distance (log) Initial Income Education Reformist Voting Large-Scale Privatization Small-Scale Privatization R 2 : Growth Regression Explanatory Variable Coefficient Estimate Error t statistic p value Deviation of Variable Quantitative Significance Constant IO Defense Price Liberalization New-Enterprise Formation R 2 : Notes: errors are heteroskedasticity consistent (White, 1980). Quantitative significance denotes the impact on the dependent variable of a one-standard-deviation increase in the explanatory variable.

6 Appendix Table A1: Regional Policy, Population Characteristics and Initial Conditions Region PRICE LPRIV SPRIV REFORM EDU INITIAL IO DEFENSE DIST 1985 From Oct-95 Dec-95 Dec :IV 1985 per 1000 Moscow ranking firms per 1000 citizens % % % workers (kms) Karelia Murmansk St. Petersburg Kaliningrad Novgorod Bryansk Ivanovo Kaluga Kostroma Moscow Oryol Ryazan Smolensk Tula Yaroslavl Mariy-El Mordovia Chuvash Kirov Nizhni Nogorod Voponezh Kursk Lipetsk Tatarstan Astrakhan Volgograd Penza Samara Saratov Ulyanovsk Adyegey Karb. Balkar North Ossetia Bashkortostan Udmurtia Kurgan Perm Sverdlovsk Chelyabinsk Novosibirsk Omsk Tomsk Tyumen Buryatia Primorskiy Khabarovsk Kamchatka Magadan Average Median Dev Minimum Maximum

7 Table 1: Growth and New-Enterprise Formation Region Geographic Territory GROWTH NEWENT Karelian Republic Northern -5.47% 4.59 Murmansk Oblast -6.10% 4.87 St. Petersburg Northwestern 1.48% Kaliningrad Oblast -2.46% 4.08 Novgorod Oblast 0.91% 4.44 Bryansk Oblast Central -9.02% 2.97 Ivanovo Oblast 0.37% 3.95 Kaluga Oblast -4.65% 5.74 Kostroma Oblast -5.23% 3.23 Moscow 15.74% Oryol Oblast -6.42% 2.52 Ryazan Oblast -7.12% 3.70 Smolensk Oblast 0.60% 2.47 Tula Oblast -2.03% 3.47 Yaroslavl Oblast -2.02% 4.89 Mariy-El Republic Volga-Vyatka -6.47% 2.09 Mordovian Republic -6.07% 1.88 Chuvash Republic -7.48% 3.09 Kirov Oblast -1.50% 1.96 Nizhniy Novgorod Oblast -2.67% 2.39 Voronezh Oblast Central Black-Earth -2.82% 3.08 Kursk Oblast -5.51% 1.71 Lipetsk Oblast -4.70% 2.40 Tatarstan Republic Volga Region -1.30% 4.23 Astrakhan Oblast -2.71% 4.18 Volgograd Oblast 2.66% 6.14 Penza Oblast -4.60% 3.91 Samara Oblast 2.35% 4.74 Saratov Oblast -5.13% 3.80 Ulyanovsk Oblast -0.97% 2.54 Adygey Republic North Caucasus -7.85% 4.00 Kabardin-Balkar Republic 0.58% 4.05 North Osetian Republic 2.28% 2.56 Bashkortostan Republic Urals -0.96% 3.12 Udmurt Republic 1.76% 3.72 Kurgan Oblast -5.02% 2.79 Perm Oblast 3.80% 3.42 Sverdlovsk 0.39% 5.08 Chelyabinsk Oblast 5.20% 3.09 Novosibirsk Oblast Western Siberia 1.93% 5.20 Omsk Oblast 0.87% 3.86 Tomsk Oblast -2.46% 4.17 Tyumen Oblast 4.65% 9.40 Buryat Republic Eastern Siberia -4.16% 3.89 Primorskiy Kray Far East -2.25% 5.01 Khabarovsk Kray -7.12% 3.69 Kamchatka Oblast -0.99% 7.06 Magadan Oblast 1.95% 7.36 Average -1.79% 4.36 Median -2.14% 3.88 deviation 4.44% 2.63 Minimum -9.02% 1.71 Maximum 15.74% 16.61