I now turn to the analysis of the patterns and determinants underlying the observed market integration results. It is hypothesized that market evolution in pre-modern China were dominated by the impact of the transport network, which was mainly composed of the navigable waterway system (Skinner, 1964; 1977a, 1977b). The importance of water transport for the grain trade during the Qing has received substantial attention in the economic history literature (Evens, 1984; Fan, 1993; Wang, 1992; Wong and Perdue, 1992; Yan and Liu, 2011). Yan and Liu (2011) attribute the higher level of market integration in South China to the region’s more developed water transport network vis-à-vis the North. In a first step to investigate the patterns and geographical determinants of market integration, I use the following linear probability regressions to analyze the N(N-1) prefecture pairs:
(4.10)
where Cij is the binary dummy for cointegration from the pairwise Engle-Granger
tests; Disijis the postal route distance in kilometers between prefectures i and j; Proij
is a dummy variable indicating whether both prefectures are locate within the same province; Riverijis a dummy variable indicating whether one of the main grain trade
rivers passes through both prefectures i and j; i and j are fixed effects for
prefectures i and j; and i*j* captures the seasonality across different cropping
i j i j ij k k ij ij k ij ij ij ij ij ij ij G Dis River Pro Dis Pro River Dis C
* * 9 1 , , 4 3 2 1 _ ) ln( ) ln(patterns. Based on Figure 2.4(b) in Section 2.5.6, i* denotes the crop area for prefecture i while j* denotes this for prefecture j.i*j*represents a group of dummies
for any combination of crop area between i* and j*, which will capture the price co- movement driven by the similar climate and harvest time in the long run35. To capture potential heterogeneity of river impact across different distance, Dis_Gij,kis
defined as a set of categorical dummies36for each distance group as below:
=1 if (k-1)×200km < Disij ≤ k×200km (k=1,2,…9) (4.11)
=0 if otherwise
The linear probability models are employed for regression (4.10) results of which are reported in Table 4.2. Results from separate regressions of equation (4.10) for rice (South) and wheat (North) will provide insights into the determinants of market integration, distinguishing markets that do from those that do not integrate. In my discussion I focus on the sign and statistical significance of covariates in the linear probability model results. Although the dependent variable is binary in nature, I prefer to employ this approach over a nonlinear estimator (e.g. Probit Model) since this make interpretation, especially of interaction effects, much easier to handle.37
In all columns reported in Table 4.2, except column 4 for Northern China, the distance variables are negative significant, which implies that the probability of market integration is higher for markets which are closer to each other. This result supports our findings in Figures 4.2 and 4.3. In column 1 and 2, the magnitudes of
35
It is possible that the agricultural price co-movement was driven by the common factor (e.g. weather) rather than market link. In this chapter, I am interested in the market determinents in the long run rather than short run such that the fixed effects of agricultural areas (discussed in Section 2.5.6) are suffiecnt to capture the climate common factors in the long run.
36
The omitted category is the group dummy for the distance between 1800km and 2000km.
37
Probit results for (4.11) also strongly support our conclusions based on the OLS results, see Appendix B.1.
k ij G Dis_ ,ln(Disij) for Southern China are smaller, which implies that in the South distance is
associated with lower trade costs since a decrease in distance between two markets ceteris paribus induces a higher likelihood of price integration. In column 4, the insignificance of the distance variable has two possible interpretations: 1) the regional market was highly integrated such that there was no association between trade and distance; 2) the regional market was highly disintegrated such that the trade costs associated with distance became infinitely large. The empirical results provided below suggest that the second explanation is a better fit to the extent of North China’s market integration.
In columns 1 and 2, both prefectures being on the same grain river significantly increases the probability of integrated markets in Southern China, but not in Northern China. This feature suggests that the network of waterways strengthen price arbitrage only in South China, a finding which is consistent with previous suggestions that waterway navigability in South China was considerably better than in the North of the country. The results for the common province dummy in the first two columns of Table 4.2 suggest that it is more likely for prefecture pairs to have integrated markets if they are located within the same province. The coefficients on the common province dummies are considerably larger than those for the common grain river dummies, which indicate that provincial borders represent comparatively stronger barriers even in a political unified economy, where currencies, weights, measures and languages were more standardized than in comparable politically fragmented regions of the world (e.g. Western Europe).38
38
It is possible that the significant province dummies are driven by the distinction between short and long distance trade rather than province border barrier since bilateral distance is much smaller within the same
This evidence of provincial barriers on grain trade implies that the water transport network on its own cannot fully explain the differences in market performance between South China and North China. It also suggests that provincial officials significantly intervened in the grain trade.
In columns 3 and 4 of Table 4.2, I add a series of interaction terms to test the additional ‘distance penalty’ on the common province and grain river dummies. In both columns, Proijis still significant to confirm the trade barrier nature of provincial
borders while the interaction term Proij× ln(Disij) reflects that the distance penalty
only applies to the provincial border in Northern China. These findings support the hypothesis that China’s market was fragmented along provincial borders, and that the extent of fragmentation was more substantial in Northern China.
In terms of water transport, the combinations between Riverij and its
interaction terms with the set of distance dummies - Riverij×Dis_Gij,k (k=1,2…,9)
capture the heterogeneous impact of waterway network across different distance group. In Table 4.3, I test the joint significance of Riverijdummy and its interaction
terms in each distance group. If I can reject the null hypothesis that their joint impact in one specific distance group (e.g. 200<x 400km) is zero (e.g. H0: + ij,2= 0,
where and ij,2 are the coefficients of Riverij and Riverij×Dis_Gij,2in regression
(4.10)), this can provide support for the significant net impact of waterways network on the market integration in this distance group.
province. As robust checks, we compare the pair cointegration percentage of within and without the same province, which yields the same qualitative conclusions. These results are in Table B.2 and B.3 in Appendix B.2.
Table 4.2 Market Integration, Geography and Institutions (OLS)
South Rice North Wheat South Rice North Wheat Dependent Variable: Cij= {0,1} (1) (2) (3) (4) ln(Disij) -0.0314 -0.0587 -0.0336 -0.0223 (0.0076)*** (0.0127)*** (0.0091)*** (0.0147) Riverij 0.043 0.0086 -0.065 0.0794 (0.0105)*** (0.0169) (0.027)** (0.091) Proij 0.222 0.264 0.343 0.833 (0.0139)*** (0.0176)*** (0.124)*** (0.146)*** Proij×ln(Disij) -0.0183 -0.096 (0.0207) (0.024)*** Riverij×Dis_Gij,1(x 200km) -0.0127 -0.0153
(0.0428) (0.0998) Riverij×Dis_Gij,2(200<x 400km) 0.1033 -0.0472
(0.0327)*** (0.0948) Riverij×Dis_Gij,3(400<x 600km) 0.1347 -0.0777
(0.031)*** (0.0928) Riverij×Dis_Gij,4(600<x 00km) 0.1607 -0.092
(0.0296)*** (0.0918) Riverij×Dis_Gij,5(800<x≤1000km) 0.1403 -0.112
(0.0285)*** (0.093) Riverij×Dis_Gij,6(1000<x≤1200km) 0.101 -0.0492
(0.0287)*** (0.0953) Riverij×Dis_Gij,7(1200<x≤1400km) 0.068 -0.117
(0.029)** (0.0945) Riverij×Dis_Gij,8(1400<x≤1600km) 0.0425 -0.127
(0.0303) (0.0941) Riverij×Dis_Gij,9(1600<x≤1800km) 0.0399 -0.1723
(0.0298) (0.0937)*
i Yes Yes Yes Yes
j Yes Yes Yes Yes
i*j* Yes Yes Yes Yes
No. of Obs. 17,030 6320 17,030 6,320
R2 0.4819 0.3865 0.4542 0.3898
Note: We omit reporting fixed effects, seasonality and the constant. Robust standard errors provided in parentheses; ***, ** and * denote 1%, 5% and 10% significance level, respectively
The results in Table 4.3 confirm that the ‘grain rivers’ had statistically significant impact on Southern market integration for distances below 1200km. The significant net coefficients between Riverij and Riverij×Dis_Gij,k in each distance
group (e.g. in 200<x 400km, -0.065+0.1033>0) suggest that Riverij in this distance
group is associated with higher probability of market integration. For South China in column 3, although the Riverijdummy enters negatively significant, its interaction
terms with distance dummies are positive significant and larger than the absolute value for the 400km to 1400km. However, for the distance group between 1200 and 1400km, the net impact from river is considered to be insignificant since the sum of coefficients is tiny (i.e. += 0.003). These results imply that water transport does not have expected positive impact for the closest pairs (≤200km) and the most distant pairs (>1200km) in South China. One possible explanation for this finding for proximate prefectures is that the distance-unrelated cost of water
Table 4.3 Heterogeneity of Water Impact on Cointegration
Distance Group Null Hypothesis p-value
South Rice North Wheat (0, 200km] H0: +ij,1= 0 0.0226 0.1218 (200, 400km] H0:+ij,2= 0 0.0527 0.2813 (400, 600km] H0:+ij,3= 0 0.0000 0.9390 (600, 800km] H0:+ij,4= 0 0.0000 0.5948 (800, 1000km] H0:+ij,5= 0 0.0000 0.2565 (1000, 1200km] H0:+ij,6= 0 0.0286 0.4191 (1200, 1400km] H0:+ij,7= 0 0.8637 0.3264 (1400, 1600km] H0:+ij,8= 0 0.2918 0.2724 (1600, 1800km] H0:+ij,9= 0 0.2652 0.0381
Note:2andij,kare the coefficients of Riverijand Riverij×Dis_Gij,kin OLS regression (11); p-
transport (e.g. cost of loading and uploading) is large relative to benefit of water transport, such that short-distance water transport was not profitable. On the other hand, the insignificant impact of river transport observed for distances larger than
1200km suggests grain river over the very long distance did not shape market
integration and its evolution in Southern China. In stark contrast, for North China, both the river and most of its interaction terms with distance dummies are insignificant. In column 4 the pattern of results for the role of river transport in northern China’s market integration strongly supports the previous conclusion that river transport has no impact on market performance in North China.
In the probit results for regression (4.10), presented in Table B.1 in Appendix B.1, I calculated the magnitude and statistical significance of the interaction effect for all interaction terms, following the re-derived formulas in Ai and Norton (2003) and Norton et.al. (2004) (see results in Appendix B.1). All results are consistent with the result pattern in Table 4.2. The Table B.4 in Appendix B.4 gives the regression of (4.10) for sub-sample excluding the possible trend stationary series, which gives the similar quantitative conclusions compared with Table 4.2. The only difference is that the river dummy for South rice is insignificant in column 1 of Table B.4, which is due to that most of possible trend stationary prefectures (have been excluded) locate in the Jiangxi Province along the largest transportable river Yangtze River.