ON THE TEMPORAL STABILITY OF GRAVITY MODELS OF INTERNAL MIGRATION by Michael J. Greenwood and Jesse Sexton* 1 Abstract--In this note, we employ 1960, 1970, 1980, 1990, and 2000 Census data on state-tostate migration in the United States to test the temporal stability of modified gravity models of migration. The double-log estimates relate to 5-year flows. The data are pooled and then stacked to allow estimation of a five-equation seemingly unrelated regression model. Little temporal work has ever been conducted on internal migration so the numerous differences that are uncovered are noteworthy. Period effects and regional effects clearly are present, and the coefficients on many variables vary considerably over time. 2 I. Introduction For many years, modified gravity models have been widely used to study both international trade patterns (Anderson, 1979) and aggregate place-to-place migration (Greenwood, 1975). With few exceptions, when applied to migration, the gravity-model approach has focused on a single period.1 Consequently, we have little idea regarding how representative the coefficients estimated in these studies are in a temporal context. Indeed, due to the paucity of temporal studies of migration, we have little knowledge regarding many temporal aspects of internal U.S. migration. The coefficients may trend, or behave cyclically, or vary from period to period in some other fashion, but the researcher has little idea regarding exactly where his estimates place in a temporal context. In this note, we develop a modified gravity model to study place-to-place interstate migration in the United States with a set of equations reflecting five periods: 1955-60, 1965-70, 1975-80, 1985-90, and 1995-2000. The estimates are double-log, which is common in the estimation of such models, and successive years are both pooled and then stacked in a seemingly unrelated regression context as we test for parameter homogeneity. Numerous differences are evident. Period and regional effects are clearly present, and the coefficients vary considerably over time, but rarely in a single direction. Although such studies are not common, a few prior studies have examined temporal changes in the determinants of migration. For example, Mueser (1989) studies five-year interstate migration over three periods (1955-60, 1965-70, and 1975-80), and White and Mueser (1994) do so over four periods (1935-40, 1955-60, 1965-70, and 1975-80). Greenwood, Hunt, and McDowell (1986) and Davies, Greenwood, and Li (2001) analyze annual data. The former paper finds some cyclicality in the migrant attractiveness of jobs and the latter finds evidence of temporal stability in many parameter estimates and some instability in others, but the studies 3 cover only 17 and 11 years, respectively, and therefore do not allow much time for changes to occur. Molloy and Wozniak (forthcoming) study several temporal aspects of U.S. internal migration. However, with few exceptions, these studies do not seek to find structural differences in their estimated models over time. II. The Model The model is a fairly standard double-log modified gravity model. The dependent variable (or migration rate) is the number of migrants five years of age and older who lived in state i five years earlier and in state j when the census was conducted divided by the at-risk population of state i (the end-of-period population of i five and over plus the out-migrants less the in-migrants). The model contains four vectors of independent variables: (1) differential economic returns; (2) costs of migrating; (3) differential amenities; and (4) regional fixed effects. The model with five equations (one for each period) is of the following form: (1) Mijt / Pit = ∑n( βint Xint) + ∑m( βjmt Xjmt) + ∑k(βijkt Xijkt) + eijt , where Mijt refers to the migration from state i to state j in period t, Pit is the at risk population of i at the beginning of the migration interval, βt refers to estimated parameters for a specific period, the X's are vectors of independent variables, ∑n refers to n variables that relate to the origin state, ∑m refers to m variables that relate to the destination state, ∑k refers to k variables that relate to specific pairs of states, and et is the error terms for each period. Comparable variables are employed for each period. Except for certain dummy variables and a distance variable, the data differ in each regression. Errors are assumed to be correlated across periods (nonzero covariances) but uncorrelated within periods. The vector for differential economic returns contains four characteristics of both state i and state j: median income (one year before the census was taken), employment growth during 4 the previous 4-year period expressed as the ratio of year t (e.g., 1994) to year t-4 (e.g., 1990), percent employment in manufacturing (t-10 at the time of the prior census), and percent of population with 16 years or more of schooling (t-10). The vector for costs of migrating contains the basic variables of the gravity model plus migrant stock: distance from state i to state j, population of i and j (both relating to at risk), and number of persons born in i and living in j at the time of the prior census. The amenities vector contains six variables for both state i and state j: percent foreign-born population (t-10), population density (t-10), January temperature, July temperature, July humidity, and a dummy variable indicating a seacoast. The vector for regional fixed effects contains dummy variables for i and j states within the following census divisions with New England as the benchmark: Middle Atlantic, South Atlantic, East South Central, West South Central, East North Central, West North Central, Mountain, and Pacific. The construction of the amenity variables is different than in previous studies and requires mention. With the exception of the percent of foreign population (t-10) and the seacoast variables, the amenity variables are built up from county-level data that are population weighted. This procedure yields a measure that is more reflective of the amenities that are experienced by the population from which the migrants are drawn or that the migrants experience at their destination. III. The Data and Econometric Approach State-to-state migration data for each year (for coterminous states) were obtained from the published Census matrices of interstate migration. The migration rate is expressed as a percentage. The total number of observations is 2,256 for each period. Means and standard deviations are reported in Table 1, except for estimation purposes the data are transformed into logarithms. The mean migration rate for each period is very similar (0.241, 0.235, 0.236, 0.232, 5 and 0.203 from earliest to latest), but drops slightly in the 1995-2000 period. Over the five periods, on average, about 23 percent of the at-risk population made an interstate move. We use lagged values or beginning-of-period values of the independent variables to avoid potential simultaneity problems. However, since the dependent variable is a specific i-j flow and the independent variable is a state characteristic, simultaneity should be less critical. Presumably, this has been the assumption when end-of-period values of the independent variables have been used in earlier studies. The econometrics proceeds in stages. First, we pool the data for all years and include dummy variables for 1970, 1980, 1990, and 2000 with 1960 as the benchmark year. To account for regional fixed effects, we also include origin and destination dummies for Census divisions with New England as the benchmark region. Next, we stack the data for consecutive years (e.g., 1960 and 1970) and use seemingly unrelated regression both to improve the efficiency of the estimates and to ascertain any differences in the values of the estimated coefficients over time.x Given the large number of observations, to get the estimates to converge we were required to stack the years two (consecutive) at a time. The climate variables seem to refer to different years. How, exactly are these measured? IV. Empirical Findings Results presented in Table 2 reflect period effects, which are clearly present, as well as regional effects, which also are clearly present. The R2 (0.91) is very high, which is common in modified gravity models. Almost all coefficients are statistically significant and most have the expected sign. The dummy variable for each year is negative, and with the exception of 1970 is statistically significant. Moreover, again with the exception of 1970, each coefficient is significantly lower than its predecessor. Other factors held constant, migration rates have fallen 6 over time, perhaps in part due to the end of the draft and also in part due later to the aging of the baby boom generation out of the most mobile age classes.1 Almost all division dummies are significant. Relative to New England, out-migration was significantly lower for Middle Atlantic, West North Central, East South Central, and West South Central, but significantly higher for East North Central, South Atlantic, Mountain, and Pacific. In-migration was lower for all divisions except Mountain and Pacific. For the most part, coefficients on other variables are highly significant and have signs that either are as expected or are not unexpected. Distance discourages migration from i to j, whereas past migrants from i to j encourages migration. Even when southern states and regional amenities are controlled, migration does not appear to have been to high-income states, a finding not uncommon in place-to-place migration studies (Greenwood, 1975). Employment growth in the destination was very attractive to migrants. The elasticity on destination employment growth (2.99) is the highest in the regression. Migration was strongly away from states with relatively high concentrations of the foreign born, which is consistent with the earlier finding of Borjas (2006), and migration rates were lower to states with high foreign-born concentrations. Higher population density discouraged both in-migration and out-migration. Both out- and in-migration rates were lower for more densely populated states. Education is a key determinant of the migration. Out-migration rates are higher from states with well-educated populations. Inmigration rates also are higher to states with well-educated populations, which was a noteworthy pattern during the last half of the 20th century. The results associated with the seemingly unrelated regression as well as tests for parameter homogeneity are reported in Table 3, where many differences are evident in the values of the estimated coefficients, although the signs tend for the most part to be consistent and as 7 expected. Only a few coefficients trend in the same direction or display no tendency to trend at all. The East North Central region regularly became less desirable as a destination. Greater employment growth in an origin always encouraged out-migration but this tendency grew somewhat over time and especially during 1995-2000. Greater percentages of college graduates at potential origins always encouraged higher rates of out-migration, but this coefficient never moved in one direction or the other. Out-migration from coastal states as origins became less likely over time, but the trend was not regular. Distance is consistently negative, and it does not trend downward in absolute value, as might be expected in light of a significantly improved national transportation system. Migrant stock is consistently positive and its coefficient varies little from period to period. States with higher incomes were significantly less likely to lose migrants during the 1955-60 and 1985-90 periods, but for the other periods the origin-income variable is not significant. The destination income variable is never significant. Destination employment growth was always strongly attractive for migrants with one of the highest estimated elasticities in each regression, but the attractiveness of jobs dipped between the 1960s and the 1970s and then rebounded during the 1980s and remained equally strong during subsequent years. Types of jobs also were important. Out-migration was regularly discouraged from states with higher percentages of jobs in the manufacturing sector, but during the first three periods studied, these same states experienced less in-migration. Reinforcement for Borjas’s finding is evident in the coefficients on the variable for foreign concentration in the origin state. This variable is positive and statistically significant for each year. Migrants in general moved out of states with high foreign-born concentrations. However, the results on the destination concentration variable are less consistent—positive and 8 significant for 1960, negative and significant for 1980 and 2000, and positive but not significant for 1970 and 1990. As a further test of the hypothesis that migrants are, on net, moving away from the foreign born, we replaced the i and j variables with the ratio (j/i), which should take a negative sign if the net effect of the foreign born is to cause migration away. The new variable is negative and significant for each year.x Again, migration tends to be lower if the destination state has a high concentration of foreign-born persons relative to the origin state. Mueser and Graves (1995) note that location-specific amenities ought to become more attractive over time as real incomes rise in all locations. However, their empirical results do not provide any support for this position, and neither do the present findings. Although the temperature variables sometimes have the expected sign, the signs on the variables are not consistent, and they do not trend in the expected direction in accord with this hypothesis. Many of the fixed regional effects are consistently significant with either a positive or negative sign. However, the Mountain and Pacific divisions are different in this respect. The Mountain division as an origin is always positive and significant until 1995-2000, when it loses its significance. As a destination, the Mountain division is positive and significant for the first two periods, insignificant for the third period, and negative and significant for the last two periods. The Pacific division as an origin is always positive and as a destination is positive for the first three periods, but not significant for the last two. Even when all of the specific variables included in the model are accounted for, the West census region appears to have become less attractive for interstate migrants. V. Summary and Conclusions The results of this study suggest that qualitatively, based on regularity of the signs on coefficients estimated for different periods, the empirical results associated with modified gravity 9 models are quite consistent and raise little cause for concern. However, the magnitude of many coefficients differ sufficiently over time for investigators to be cautious in the weight they place on specific coefficients. Footnotes *University of Colorado. 1. Rosenbloom and Sundstrom (2004) study U.S. interstate migration over a long span of history (1850-1990), but they analyze microdata and do not estimate modified gravity models. They conclude that interstate migration propensities fell during the last half of the 19th century and then rose until about 1970. They attribute the rise during the 20th century to increased educational attainment. 2. Molloy and Wozniak (forthcoming) address the issue of declining U.S. internal migration rates but come to no strong conclusions regarding the causes. References Anderson, James E., "A Theoretical Foundation for the Gravity Equation," American Economic Review, 69 (March, 1979), 106-116. Borjas, G. J., "Native Internal Migration and the Labor Market Impact of Immigration," Journal of Human Resources, XLI (XXXX, 2006), 221-258. Davies, Paul S., Michael J. Greenwood, and Haizheng Li, "A Conditional Logit Approach to U.S. State-to-State Migration," Journal of Regional Science, 41 (May, 2001), 337-360. Graves, Greenwood, Michael. J., “An Analysis of the Determinants of Geographic Labor Mobility in the United States,” The Review of Economics and Statistics, LI (May 1969), 189-194. Greenwood, Michael. J., “Research on Internal Migration in the United States: A Survey,” Journal of Economic Literature, 13 (June, 1975), 397-433 10 Greenwood Michael J., Gary L. Hunt, and John M. McDowell, "Migration and Employment Change: Empirical Evidence on the Spatial and Temporal Dimensions of the Linkage," Journal of Regional Science, 26 (May, 1986), 223-234. Molloy, Raven, Christopher L. Smith, and Abigail Wozniak, "Internal Migration in the United States," Journal of Economic Perspectives, 25(Spring, 2011), 1-42. Molloy, Raven, and Abigail Wozniak, "Labor Reallocation over the Business Cycle: New Evidence from Internal Migration," Journal of Labor Economics (forthcoming). Mueser, Peter, "The Spatial Structure of Migration: An Analysis of Flows in the USA over Three Decades," Regional Studies, 23(June, 1989), 185-200. Mueser, Peter.R., and Philip E. Graves, "Examining the Role of Economic Opportunity and Amenities in Explaining Population Redistribution," Journal of Urban Economics, 37(1995), 176-200. Rosenbloom, Joshua, L., and William A. Sundstrom, "The Decline and Rise of Interstate Migration in the United States: Evidence from the IPUMS, 1850-1990." In Research in Economic History, v.22, ed. Alexander Field. Amsterdam: Elsevier, 2004, 289-325. White, Michael J., and Peter R. Mueser, "Changes in the Demographic Determinants of U.S. Population Mobility: 1940-80," Review of Regional Studies, 24(Winter, 1994), 245-264. 11 Table 1 Summary Statistics: Means & Standard Deviations Variable Migration Rate i → j Distance i → j Population (thousands) Migrant Stock i → j Median Family Income Employment Growth Pct. of Jobs in Manufact. Pct. with College Degree Pct. Foreign-Born Population Density January Average Temp. July Average Temp. July Average Humidity Variable Migration Rate i → j Distance i → j Population (thousands) Migrant Stock i → j Median Family Income Employment Growth Pct. of Jobs in Manufact. Pct. with College Degree Pct. Foreign-Born Population Density January Average Temp. July Average Temp. July Average Humidity Pooled Mean Std. Dev. 0.229 1036.6 4325.7 23170.1 43362.3 1.1 23.369 6.910 4.308 1048.4 31.9 75.1 55.4 0.447 584.1 4681.3 55513.2 9279.7 0.1 10.571 3.772 3.719 2285.1 11.3 4.9 15.6 1980 Mean Std. Dev. 0.236 1036.6 4338.2 22238.1 45615.1 1.153 24.101 5.539 3.298 974.7 31.9 75.1 55.4 0.399 584.2 4506.0 52538.6 5332.4 0.077 9.128 1.084 2.712 2003.9 11.3 4.9 15.6 1960 Mean Std. Dev. 0.241 1036.6 3280.3 15318.0 31291.3 1.1 29.205 3.303 5.442 1160.1 31.9 75.0 55.3 0.603 584.2 3306.8 38244.8 5735.7 0.1 12.898 0.814 4.373 2729.2 11.3 4.8 15.6 1990 Mean Std. Dev. 0.232 1036.6 4754.8 28167.5 47111.0 1.043 20.256 9.204 4.105 963.8 32.0 75.1 55.4 0.407 584.2 5060.0 63100.7 7922.4 0.054 7.404 1.716 3.322 1928.9 11.3 4.9 15.6 1970 Mean Std. Dev. 0.235 1036.6 3839.1 19276.3 42451.1 1.1 26.906 4.039 4.076 1099.3 31.9 75.1 55.4 0.447 584.2 3993.4 47735.1 6399.8 0.1 10.835 0.871 3.227 2435.7 11.3 4.9 15.6 2000 Mean Std. Dev. 0.203 1036.6 5416.0 30850.4 50343.0 1.069 16.377 12.463 4.619 1044.2 32.0 75.1 55.3 0.331 584.2 5841.1 69058.0 7153.2 0.061 5.597 2.527 4.329 2230.1 11.3 5.0 15.6 12 Table 2 Pooled Interstate Migration: Logarithmic Regression Coefficients & Absolute t-Values Variable β Costs of Migration Distance i → j -0.4689 Population i -0.5550 Population j 0.4487 Migrant Stock i → j 0.5775 t (41.87) (51.97) (42.15) (90.30) Variable 2000 1990 1980 1970 β Yearly Fixed Effects -1.3207 -0.8557 -0.5460 -0.0195 t (20.00) (16.30) (16.43) (0.75) Pct. Foreign-Born i Pct. Foreign-Born j Population Density i Population Density j January Average Temp. i January Average Temp. j July Average Temp. i 0.1427 -0.0407 -0.0464 -0.0343 -0.0953 -0.1759 2.2887 (15.46) (4.37) (7.75) (5.72) (2.92) (5.38) (10.70) Census Division Fixed Effects & β0 Middle Atlantic i -0.2419 (9.67) Middle Atlantic j -0.5054 (20.16) East North Central i 0.1226 (4.67) East North Central j -0.4219 (16.02) West North Central i -0.1146 (3.90) West North Central j -0.3510 (11.90) South Atlantic i 0.0938 (4.04) South Atlantic j -0.2354 (10.13) East South Central i -0.1611 (5.19) East South Central j -0.4313 (13.89) West South Central i -0.1510 (4.82) West South Central j -0.4304 (13.69) Mountain i 0.4403 (10.55) Mountain j 0.0343 (0.81) Pacific i 0.6192 (14.57) Pacific j 0.0721 (1.65) β0 -18.5372 (11.28) July Average Temp. j July Average Humidity i July Average Humidity j Coastal i Coastal j 1.8431 0.0051 0.1658 0.1493 0.0291 (8.63) (0.14) (4.41) (10.51) (2.05) Observations: r-squared: Economic Differentials Median Family Income i -0.5407 Median Family Income j -0.0733 Employment Growth i 1.2751 Employment Growth j 2.9862 Pct. of Jobs in Manufact. i -0.1418 Pct. of Jobs in Manufact. j -0.0591 Pct. with College Degree i 0.7503 Pct. with College Degree j 0.0922 (9.08) (1.24) (14.92) (34.78) (9.88) (4.12) (18.09) (2.20) Amenities 11280 0.9109 13 Table 3 Interstate Migration: Seemingly Unrelated Logarithmic Regression Coefficients, Absolute z-Values, and Sig. Differences 1960 Variable β 1970 z β 1980 z Distance i → j Population i Population j Migrant Stock i → j -0.5579 -0.5123 0.4056 0.5175 (21.25) ≫ (22.21) (17.60) ≪ (39.10) -0.4866 -0.5530 0.4695 0.5335 (22.00) (26.18) (22.46) (43.86) Median Family Income i Median Family Income j Employment Growth i Employment Growth j Pct. of Jobs in Manufact. i Pct. of Jobs in Manufact. j Pct. with College Degree i Pct. with College Degree j -0.3379 -0.2508 0.9061 2.3453 -0.3226 -0.1565 0.4791 -0.1155 (2.28) (1.70) (4.24) (10.92) (8.85) (4.28) (4.90) (1.18) -0.2181 -0.2091 1.3497 1.0571 -0.1560 -0.0687 0.5343 0.1747 (1.40) (1.35) (6.87) (5.34) (5.16) (2.27) (6.31) (2.05) Pct. Foreign-Born i Pct. Foreign-Born j Population Density i Population Density j January Average Temp. i January Average Temp. j July Average Temp. i July Average Temp. j July Average Humidity i July Average Humidity j Coastal i Coastal j 0.1446 0.1086 -0.0465 0.0038 0.4010 -0.1719 0.3236 4.2170 -0.0316 0.6944 0.1441 0.2099 (6.11) (4.49) (3.75) (0.31) (5.24) (2.21) (0.59) (7.71) (0.37) (8.21) (4.44) (6.49) 0.2277 0.0209 -0.0505 -0.0088 -0.0942 -0.3897 2.8131 4.0619 0.3631 0.3224 0.1867 0.0344 (10.79) (0.96) (4.46) (0.77) (1.46) (5.96) (6.43) (9.30) (4.89) (4.35) (6.89) (1.27) < ⋙ ⋙ ≫ ≪ ⋘ ⋙ ⋙ ⋘ ⋘ ⋘ ⋙ ⋙ β ⋘ ⋙ ⋘ ⋙ ⋘ ≫ ≪ ≫ ⋘ ⋘ ⋘ > ⋙ ≫ ⋙ ≫ 1990 z β 2000 z β z -0.6249 -0.4252 0.6394 0.4560 (29.48) (18.78) (28.85) ⋙ (39.76) -0.6487 -0.3907 0.5622 0.4798 (31.73) ⋙ (19.43) < (27.37) (38.07) ≪ -0.5818 -0.4291 0.5745 0.5110 (28.05) (21.18) (26.91) (39.63) 0.0428 0.0211 1.5611 2.8361 -0.0709 -0.1509 0.6666 0.2310 (0.31) (0.15) (6.63) (11.97) (2.04) (4.35) (6.94) (2.40) ≪ -0.2884 -0.0360 1.2704 3.0337 -0.1710 -0.0421 0.7052 -0.1111 (2.63) ≫ (0.33) (6.20) ⋘ (14.71) (5.71) (1.40) (7.61) (1.19) ⋘ 0.0214 -0.1265 3.1135 2.7058 -0.1679 -0.0438 0.7081 0.2885 (0.18) (1.08) (11.02) (9.56) (5.47) (1.43) (7.88) (3.20) 0.1675 -0.1148 -0.0367 -0.0684 -0.3474 -0.2456 3.4388 1.0140 0.1512 -0.1451 0.1150 0.0503 (6.59) (4.49) (2.51) (4.68) (4.73) (3.34) (7.34) (2.15) (1.83) (1.76) (3.64) (1.59) 0.1341 -0.0819 -0.0934 -0.0453 -0.0915 -0.0960 2.2056 1.6708 -0.1106 -0.0861 0.0354 -0.0029 (5.20) (3.16) (6.80) (3.24) (1.55) (1.62) (5.84) (4.44) (1.37) (1.06) (1.28) (0.11) ⋘ ⋙ ⋙ ≫ 0.09226 (3.32) ⋙ 0.00480 (0.17) ⋘ -0.09819 (7.64) -0.06713 (5.17) ⋙ -0.14076 (2.25) ⋙ -0.03013 (0.48) ⋙ 1.57516 (3.79) 1.28841 (3.09) 0.06226 (0.88) ≪ -0.32723 (4.59) 0.13415 (4.90) ⋘ -0.05037 (1.84) ⋘ ≪ ⋙ ⋙ > 14 Table 3 (cont.) Interstate Migration: Seemingly Unrelated Logarithmic Regression Coefficients, Absolute z-Values, and Sig. Differences 1960 Variable Middle Atlantic i Middle Atlantic j East North Central i East North Central j West North Central i West North Central j South Atlantic i South Atlantic j East South Central i East South Central j West South Central i West South Central j Mountain i Mountain j Pacific i Pacific j β0 Observations: Pseudo r-squared: 1970 β z -0.3183 -0.5847 0.1187 -0.1651 -0.0813 -0.2470 0.1327 -0.1929 -0.0980 -0.3539 -0.1471 -0.4264 0.2813 0.7684 0.4166 0.6359 -19.8358 (5.49) (10.09) (1.98) (2.74) (1.17) (3.54) (2.21) (3.19) (1.25) (4.51) (1.87) (5.36) (2.93) (7.88) (4.26) (6.31) (4.57) 2256 0.9159 ≫ ⋘ ≫ > ⋘ ⋙ ⋘ > ≪ 1980 β z β -0.1719 -0.6549 0.2149 -0.3917 0.0783 -0.3713 0.2165 -0.1728 0.0396 -0.3667 -0.0276 -0.3645 0.8481 0.2115 0.8646 0.4446 -30.0888 (3.51) (13.42) (4.20) (7.61) (1.26) (5.92) (4.46) (3.55) (0.60) (5.51) (0.40) (5.31) (10.07) (2.47) (10.05) (5.04) (7.91) -0.2280 -0.5080 -0.0076 -0.5253 -0.2443 -0.3209 0.0167 -0.2658 -0.2349 -0.4133 -0.3326 -0.1643 0.3448 0.0087 0.6028 0.2108 -24.8448 2256 0.9335 ≫ ⋙ ≪ ⋘ ⋙ ⋘ ⋘ ≫ ⋙ > ≫ ≫ 1990 z (4.02) (8.95) (0.12) (8.37) (3.43) (4.51) (0.29) (4.58) (3.03) (5.32) (4.32) (2.13) (3.47) (0.09) (6.28) (2.11) (6.78) 2256 0.9029 β ⋘ ⋙ ⋙ ⋙ ⋙ ⋘ ⋘ ⋘ ⋙ 2000 z -0.2310 -0.5894 0.1813 -0.4534 0.1392 -0.3948 0.0634 -0.1436 -0.0689 -0.4385 0.2980 -0.6631 0.7105 -0.3938 0.6439 0.0755 -12.9818 (4.67) (11.9) (3.47) (8.62) (2.17) (6.11) (1.27) (2.87) (1.02) (6.48) (4.28) (9.49) (8.33) (4.60) (8.19) (0.94) (4.37) 2256 0.9350 ≪ ⋘ < ⋘ ⋘ ⋘ ⋘ ⋙ ≫ ≫ < ⋘ β z -0.3454 -0.5235 -0.2784 -0.5484 -0.3761 -0.4361 -0.0807 -0.2224 -0.3782 -0.5436 -0.3990 -0.5705 0.0266 -0.1817 0.4886 -0.0771 -21.5410 (7.29) (10.98) (5.43) (10.74) (6.50) (7.66) (1.72) (4.74) (6.11) (8.78) (6.50) (9.34) (0.33) (2.28) (6.36) (0.99) (8.98) 2256 0.9324 15