Chinese property stocks genuinely repriced on 29 September 2026, but Pan establishes nothing tradable beyond that move. Accounting for correlation cuts the BMP z-statistic 4.8-fold, from 10.48 to 2.17. The subsequent fade might invite a reversal trade. It also spans a seven-day market closure.
The subsidy and the share-price response
Caijin [2026] No. 95 pays one percentage point a year of a first-time buyer's mortgage interest for up to five years. Eligibility is confined to new commercial loans on homes no larger than 120 square metres and costing no more than RMB 1.5 million. Subsidised principal has a cap of RMB 1 million. Central and local government divide the cost 90:10, and the programme is provisional for one year. A household can save about RMB 50,000 over five years at most. Pan argues that developer equity could not plausibly capitalise those cash flows directly. We agree. His explanation instead rests on expected demand and the signal of further support.
The sample comprises the 69 constituents of the CSI All-Share Real Estate Index (931775), the broad A-share property benchmark. Shanghai accounts for 36 and Shenzhen for 33. For each firm, Pan estimates a single-factor market model against the CSMAR value-weighted A-share return, using 240 days ending eleven sessions before the event. Mean beta comes to 0.813. He sets the event day at 29 September; markets then closed from 1 to 7 October.
Average abnormal return on the event day is +3.81%. Returns were positive for 94.2% of the 69 firms, with a median of +3.34%. Across market values, the gain amounts to RMB 32.57 billion, about 4.1% of a RMB 793.0 billion base. The surrounding sessions complicate the picture. The five before the event contribute +4.16% (t 6.10), of which +3.39% arrives on 21 September alone. The five afterwards lose 1.82%. Cumulative abnormal return reaches +8.51% on 30 September before finishing the window at +6.14%. Pan interprets the sequence as advance pricing of a wider policy cluster, then a partial reversal.
How much evidence can one sector-wide event supply?
There is one date and one sector-wide shock. Estimation-window residuals have an average pairwise correlation of 0.245. The Boehmer-Masumeci-Poulsen (BMP) statistic standardises each firm's abnormal return and accommodates event-induced variance, while treating firms' abnormal returns as independent. The Kolari-Pynnönen adjustment addresses that dependence. Applying it reduces BMP from 10.48 to 2.17. Pan reports the reduction himself and says the result remains significant at 5%. His abstract nevertheless features 10.67, 12.90 and 10.48, leaving out 2.17.
The placebo gives a more useful comparison. Repeating the event-day calculation for each of the 240 estimation days, Pan obtains a standard deviation of 1.24 and a range from -3.13 to 3.59. The event's +3.81% exceeds them all, clearing the maximum by 0.22 points. A return of 3.81 against a placebo standard deviation of 1.24 puts the announcement about 3.1 standard deviations out. By contrast, +3.39% on 21 September falls within the placebo range. This test would therefore have missed a day almost as large as the announcement.
Two further checks supply no independent evidence of significance. The equal-weighted portfolio of 69 abnormal returns is itself the average abnormal return; agreement "to four decimals" follows arithmetically. Breadth of 94.2% is also consistent with a common sector shock when residual correlation is 0.245. We did not find the Corrado rank result in the table of checks, nor the listing-status and suspension checks cited in the conclusion. The table contains the baseline, winsorisation, the Kolari-Pynnönen adjustment and the placebo.
Pan also reports what happens when he shifts the event date. Appendix C moves day zero to 30 September and finds a one-day effect of +0.55% (t 1.58). He takes this as evidence that the response occurred on 29 September. Under the shifted convention, the 3.81% returns within the [-1,+1] window, whose gain is +3.98%. The objection is circularity. Pan describes the one-day window as "chosen ex ante", yet defines day zero as "the session in which the notice became public and in which... the sector recorded its anomalous return". That second criterion uses the outcome to select the date. We found no intraday publication time. If publication came after the close, the 3.81% accrues to whoever knew early.
Small errors carry weight in a one-event paper. The text calls the [-2,+2] window's +3.90% smaller than the +3.81% event-day return. It also blames a day (t=+4) outside that window.
A 1.82% fade across the National Day closure
Post-event returns sum to -1.82% (t -4.13), including -2.16% on 12 October. Even so, the full eleven-day window finishes at +6.14%, retaining about 72% of the +8.51% peak. The paper describes this as only partly durable. Shorting at the 30 September close, near the +8.51% peak, would mean holding through the 1 to 7 October closure. Across the four sessions from 8 to 13 October, cumulative abnormal return drops from +8.51% to +6.14%, a decline of 2.37 points. These are market-model abnormal returns from one in-sample event. Costs and borrow would change a short's P&L.
Pan himself names alternative explanations: holiday transaction data, follow-on announcements and mean reversion. The window ends on 14 October. Yet the abstract settles on a reversal "once the subsidy's modest scope was digested". The body never tests that account against those three alternatives.
Price limits raise a separate question. Event-day abnormal returns for Huafa, Vanke, Cinda and Binjiang range from 9.96% to 10.19%, on boards subject to 10% daily limits. Because abnormal returns differ from raw returns, we cannot establish whether those limits bound. If they did, some demand from 29 September would have spilled into 30 September, when sector AAR was +0.54%. We did not find any discussion of limits.
Attribution still matters
Pan acknowledges that "the entire event-day jump therefore cannot be attributed to the subsidy alone". The PBoC/NFRA residential-credit Opinion and the completed-unit sales Notice appeared in the same quarter. He also acknowledges the absence of a control group. His abstract nevertheless concludes that "even a fiscally small, targeted measure can move sector valuations materially, but only partly durably". Qualifying durability leaves attribution unresolved. His concession weakens the "targeted measure" part of the claim.
The fiscal comparison also lacks a denominator. RMB 32.6 billion would cover the maximum five-year subsidy for about 650,000 households. Pan calls the outlay small "even under generous assumptions", without supplying a borrower count. Without that count, the order-of-magnitude claim has no support.
Tefa Service, the biggest gainer at +13.45%, is a property-service firm. Pan says the claim that developers serving eligible segments led the move has not been formally tested.
We could not replicate this.
Our data contain neither Shanghai and Shenzhen equities nor the 931775 constituent snapshot, and we have no timestamped record of Chinese housing-policy releases. Substituting a US real-estate proxy would test exposure to US rates, a different hypothesis.
The 29 September repricing is real at roughly three sigma. Pan's own proposed extension would change our view: rank the cross-section by exposure to sub-120-square-metre, sub-RMB-1.5-million product, then show that this exposure predicts the 3.81%.