1.6.4.1 Benchmark Result
Table 1.10 reports the baseline results of estimating equation (1.3) using cumulative abnormal returns from day -20 to day 5 (CAR[−20,5]) as the dependent variable. Bootstrapped standard errors are reported. Stock exchange dummies are included in all specifications to capture common factors pertaining to the stock exchanges. Columns (1) and (2) represent the unconditional effects coming from the firm traits of interest – non-residential real estate concentration, geographic exposure of housing properties and degree of political connections, which alone explain more than 17% of the variation in abnormal stock returns. The preferred benchmark specification in column (3) controls for all other firm covariates, such as firm size, leverage and profitability. This weakens but does not wipe out the significance of the estimates of interest.
As expected, the coefficient on non-residential real estate is positive and signifi- cant. Namely, a concentration in non-residential properties brings a higher abnormal return of 3%, buffering the negative shock from the housing policy. Residential ex- posure in speculative markets causes more negative abnormal returns but the effect is barely significant in a statistical sense. However, this should not be interpreted as geography not mattering. In fact, the economic importance of geographic expo- sure is likely to be obscured by the fact that quite a few real estate firms have been diversifying their housing properties across markets. For example, more than 50% of the firms in the sample operate in more than three provinces/municipalities, and most of them have been expanding their business to provincial cities and second-tier or third-tier cities rather than confining themselves to first-tier cities.
The coefficient on the central state-controlled firms is -0.054 and statistically sig- nificant, suggesting that firms controlled by the central government or other central
state-owned enterprises suffer an additional abnormal return of -5.4% compared to private real estate firms over the [−20,5] event window. The response of local state- controlled firms is not distinguishable from that of private firms. It is also worth noting that the relative ranking of coefficients on political connections is robust to the inclusion of other firm-level characteristics. On average, central state-controlled firms suffer the most severe losses and private firms the least.
The negative coefficients on state-controlled firms suggest that the second channel of political connections (as discussed in Section 1.6.1.3) plays a dominant role in this case. Namely, the expectation that state-controlled firms would suffer disproportion- ally from a reduction in government support and that they would follow government policies more actively, outweighs the widely-acknowledged benefits of political connec- tions in buffering negative market shocks. In addition, the second channel is assumed to affect firms affiliated with the central government more directly because the hous- ing policy is decided by the central government. This is born out in the empirical results where state-controlled firms affiliated to the central government experience more negative abnormal returns than those affiliated with local governments.
Column (4) uses the percentage of state-controlled shares held by the largest shareholder as a proxy for firm political connections. I also include its interaction term with being a central state-controlled firm to gauge the difference between con- nections with the central government versus local governments. In spite of its flaw in reflecting true variations in political connections across firms (as discussed in Section 6.1.3), this continuous state-share variable predicts more negative abnormal returns for state-controlled firms, especially for those connected with the central govern- ment.57 However, this effect is not statistically significant at conventional levels.
Coefficients on other firm-level covariates are suppressed in the table. Only firm 57To interpret the coefficient, a one percentage point increase in the state shares held by the largest
shareholder would lead to an additional -10 basis points abnormal return for a central state-controlled firm.
size has a significant impact. A 1% increase in market capitalization leads to an additional abnormal return of around -1.5 basis points. In case of outliers driving the result, column (5) repeats the same regression as column (3) but on a restricted sample where firms in the top and bottom 2% distribution ofCAR[−20,5] are trimmed. This produces similar estimates as the full sample.
Following discussions in Section1.6.3, I then capture omitted factors by controlling for firms’ growth rate in terms of market capitalization and total asset (at book value) from 2008 to 2009. Estimates are reported in column (2) and (3) of Table 1.11, which shift the baseline estimates in column (1) only slightly. Firm concentration in non-residential business and political connections continue to matter beyond omitted factors. In addition, the coefficient on the 2008-to-2009 growth rate of total assets is significantly negative, suggesting that all else equal, an additional 1% increase in the book value of total asset during the boom is associated with an additional cumulative abnormal return of -4.2 basis points.
1.6.4.2 Robustness Tests: Alternative Event Windows
In order to confirm that the above results reflect the true relationship between ab- normal policy response and firm traits of interest rather than some spurious local correlations in data, I estimate equation (1.3) using cumulative abnormal returns over alternative event windows as the dependent variables in Table 1.12.
Column (1) is the benchmark result (taken from column (3) in Table 1.10) for comparison. Columns (2) to (3) use more immediate abnormal returns after the release of the official policy document, CAR[−20,3] and CAR[−20,4], respectively as dependent variables. Both of them produce similar results to the benchmark. Firms with non-residential real estate concentrations suffer less compared to housing- oriented developers. Similarly, central state-controlled firms exhibit significantly more negative abnormal returns than private firms, and the discrepancy is bigger when we
look at more instantaneous responses. Column (4) examines longer-run abnormal return until day 10 and the relationship is more or less similar to that in the short run. The benefit of engaging in non-residential real estate seems to accumulate 10 days after the policy announcement. The last column uses alternative starts of event windows and examines CAR[−10,5]. Again, the estimates stay similar to those in the benchmark specification.
1.6.4.3 Result over the Course of Policy Release
The next question of interest involves how the relationship between abnormal re- turns and firm traits changes over the course of the policy announcement. To in- vestigate this, I estimate the baseline specification sequentially using CAR[−20, k] (k ∈ [−20,20], k ∈ Z) as the dependent variables. Figure 1.10 and 1.11 plots the point estimates and the 95% confidence intervals over k for the four coefficients of interest. The figures are truncated atk = 10 because this paper focuses on short-run stock responses. Relationships and abnormal returns over the longer run are subject to noise.
Figure1.10 plot coefficients on non-residential real estate concentration and spec- ulative markets, respectively. Again, property diversification is an important deter- minant of abnormal response to the housing policy shift. Compared with housing- oriented developers, firms with non-housing concentration enjoy an additional CAR of 2% to 3%, mitigating the negative policy shock. Residential exposure in superstar markets, however, does not matter significantly in statistical sense. As is discussed above, those firms have already diversified their portfolios geographically. Figure1.11 plot the two coefficients on state-controlled firms. Central state-controlled firms suf- fer an additional CAR of -5% to -7% compared to private firms during the course of the policy issuance. The coefficient on local state-controlled firms, however, is not significantly distinguishable from zero.
One noteworthy finding is that firm traits begin to matter only after we observe abnormal returns. This is consistent with the intuition that the relationship between abnormal returns and firm traits should appear only when the abnormal returns show up. During normal periods, there is little justification for such a relationship. The effect of political connections synchronizes the time line of the policy announcement, which does not become significant until people learned of the BFA speech on mortgage regulations (after day -4). The value of non-residential real estate concentration, however, is not revealed until the full release of the official policy document (after day 1). This timing difference can be explained by how the policy content was gradually released. Before day 2, only the core piece of “Document Number 10” – tightening mortgage requirements for speculators – was unveiled to the public. Only after reading the entire policy document did investors learn that the policy was only targeted at the housing market, thus allowing them to form differential beliefs about firms with non-residential real estate concentrations.