For Vietnamese equity traders watching overnight crypto selloffs, most of the associated return comes after the opening bell. A close-to-close regression conceals the timing. Pham and Phan estimate the VN-Index response to an eight-coin overnight crypto downside measure across 1,489 sessions, from 12 August 2020 to 31 July 2026. They report a daily coefficient of -17.2 bp per one-standard-deviation increase (SE 5.4, Holm p 0.004). Only about a seventh falls in the opening gap: -2.3 bp (SE 1.9, p 0.230). The open-to-close coefficient is -14.9 bp (SE 4.4).
The intraday share is therefore 0.86, with a delta-method interval of 0.68 to 1.05 and a block-bootstrap interval of 0.68 to 1.14. Singapore reverses that allocation. For the Straits Times Index, the opening coefficient is -7.2 bp and the intraday coefficient is -1.7 bp, giving an intraday share of 0.19. Both STI coefficients remain negative. The STI's opening-minus-intraday difference is -5.5 bp (95% CI -10.4 to -0.5). Its raw p of 0.031 rises to a Holm p of 0.092, while the share interval extends from -0.08 to 0.47. This split resembles the US evidence in Song (2026), where overnight Bitcoin losses register mainly at the open. Malaysia's KLCI divides the response into -1.8 and -2.1 bp, although its daily total (-3.9 bp, Holm p 0.160) is indistinguishable from zero. Indonesia's JKSE records -9.8 and -6.8 bp.
Building the overnight measure
The authors construct the signal for each equity session using five-minute bars from eight Binance spot pairs quoted in USDT: BTC, ETH, XRP, ADA, SOL, DOGE, LTC and TRX. They measure returns from the previous local equity close to a pre-open cutoff. In Vietnam, the window starts at 14:45 the prior day and ends at 08:45 local time. Their measure, CMDS, multiplies the share of the eight coins that declined by the absolute median loss among the decliners. Two coins falling 4% produce a lower score than seven falling 3%. Each market's measure is standardised separately.
A broad, deep crypto selloff can thus be observed before Ho Chi Minh City opens. The regressions include six US daily controls (S&P 500, Nasdaq, VIX change, dollar index, 10-year yield change, WTI), along with the lagged index return and weekday and month-year fixed effects. The authors use identical sessions for all three return outcomes. That makes the opening and intraday coefficients sum exactly to the daily coefficient. They describe the allocation with D, the opening coefficient minus the intraday coefficient; positive D means the intraday coefficient is more negative.
For the VN-Index, D reaches 12.5 bp (95% CI 4.3 to 20.7, Holm p 0.014). The corresponding figures for VN30, VNMidcap and VNSmallcap are 12.7, 17.4 and 16.8 bp. On common trading dates across the full sample, Vietnam's D exceeds Singapore's by 17.3 bp (CI 7.0 to 27.6, Holm p 0.003, N 1,455). The difference from Malaysia is 12.8 bp (Holm p 0.006). Against Indonesia, the contrast is 13.3 bp, with Holm p 0.059.
These are conditional associations. Month-year fixed effects turn the coefficients into within-month comparisons, and the paper acknowledges that profitability would also depend on forecasts, liquidity and transaction costs. Removing the six US controls increases the coefficients' magnitudes. Those controls absorb about half of Singapore's uncontrolled coefficient (-17.7 to -8.9 bp), compared with about a quarter for the VN-Index (-23.7 to -17.2 bp).
We could not backtest this. A backtest requires actual Vietnamese, Singaporean, Malaysian and Indonesian index opening and closing records aligned to local sessions. Our data contain none of those records. A US-listed Vietnam fund cannot substitute for them: it trades while Ho Chi Minh City is closed and cannot reproduce a decomposition tied to the local opening print.
Could stale Vietnamese openings explain the split?
Stale opening prints are the obvious challenge. When the reported index open repeats the previous close, the measured gap is zero. Any repricing then appears intraday, producing exactly the Vietnam pattern. The authors define stale openings as those within 0.5 bp of the prior close. Their frequency ranges from 5.6% to 11.4% across the Vietnamese series, versus 1.3% for the STI, 1.6% for the KLCI and 26.1% for the JKSE.
The paper subjects the finding to several opening-price checks. After stale sessions are removed, Vietnamese D values range from 11.9 to 17.2 bp. Excluding months with clusters of stale opens produces 12.0 to 20.0 bp. Replacing the main provider's openings with SSI iBoard observations where the sources disagree gives 12.2 to 17.2 bp. Using KBSV openings yields 12.4 bp across the full 1,489 sessions and 12.6 bp across 1,406 non-stale sessions.
The data concern remains substantial. As the authors disclose, a provider correction to opening prices reduced VN-Index D from 18.7 to 12.5 bp. The three cross-market contrasts also fell, from 23.4, 19.1 and 21.4 bp to 17.3, 12.8 and 13.3 bp. That correction removed a third of the VN-Index contrast. And the 0.5 bp filter detects only fully stale openings. An index can pass it even when some constituents still carry yesterday's prices, leaving part of the repricing to appear during the session. Constituent prices would resolve that question; the authors include them among the data that could explain the timing differences.
Other explanations receive two checks, neither of which tests partial staleness. Across 12 Vietnamese equations, one-session lead coefficients of CMDS have p of at least 0.215, so CMDS does not proxy for the following session. Including in-session regional index returns and in-session BTC-ETH returns leaves D at 96% to 103% of its matched baseline. Contemporaneous moves therefore do not absorb the difference.
A five-analysis penalty raises Vietnam's p-values to 0.061-0.083
The authors acknowledge choosing the timing focus after examining five analyses of the same data: conditional means, extreme losses, lower quantiles, closure-length interactions and timing. For their sensitivity check, they multiply the Holm-adjusted p-values of the four Vietnamese allocation differences by five. The resulting range is 0.061 to 0.083. In their defence, an earlier analysis version had specified the within-session decomposition and the three primary VN-Index contrasts. They explicitly describe the fivefold figures as sensitivity values. That disclosure is clear, and those are the numbers to carry forward.
The stability checks draw on the same sample, as the authors also acknowledge. Through 3 August 2023, the first half produces D of 12.5 bp (744 sessions, p 0.018) and an intraday share of 0.94. In the second half, D is 11.9 bp (745 sessions, p 0.084), with a share of 0.77. The paper runs no equality test. A signed BTC-ETH measure that also counts gains preserves the ordering: 3.5 bp at the open and 12.2 bp intraday. Its difference is -8.7 bp (p 0.011).
The effect is small: the -14.9 bp intraday coefficient compares with a 1.245% daily VN-Index standard deviation.
Across all four Vietnamese indices, the opening coefficient remains small and the intraday coefficient larger through four separate opening-price checks. That ordering is the paper's most credible finding. Its standing as a tradeable warning signal is weaker: the association remains in-sample and loses 5% significance after accounting for selection. A D near 12 bp on sessions after July 2026, calculated from constituent-level opening prices, would change our view.