Do auditor changes, director resignations and defaults predict stock returns in India?
Short answer: On 4.5 years of NSE Reg-30 announcements, no red-flag category predicted returns once compared with same-day peers.
Published 2026-10-30 · original backtest on Indian market data by TRINETRA
Date: 2026-10-30 Scripts: engine/scripts/bench_redflag_events.py, engine/scripts/bench_redflag_robust.py Data: NSE corporate announcements 2022-01 → 2026-07, 55 monthly pulls, cached in data/audit/redflag_announcements.json. Results in BENCH_redflag_events.json and BENCH_redflag_robust.json. Idea source: trader-skill trader-news, which ranks 8-K items 4.01 (auditor change), 4.02 (non-reliance), 5.02 (officer change) and 2.04 (debt acceleration) as IMPORTANT.
Verdict: no category predicts forward returns. One false positive was caught and retracted.
Stage 1: same-stock placebo, Benjamini-Hochberg over 18 cells
| category | priced events | 20d median excess | 60d | 120d | best q |
|---|---|---|---|---|---|
| Change in Auditors | 1,013 | −0.04 (null −0.19) | −0.19 (+0.36) | −2.10 (+1.56) | 0.018 |
| Independent director resigned | 396 | +0.65 | +1.01 | +3.89 | 0.49 |
| Debt default | 106 | −1.73 | +5.91 | +14.24 | 0.49 |
| Fraud/default/arrest | 23 | −0.46 | −4.01 | +3.01 | 0.82 |
| Insolvency (CIRP) | 20 | −0.67 | +7.62 | +11.24 | 0.49 |
| Rating upgrade | 42 | −1.79 | −2.90 | +1.45 | 0.49 |
The rating-downgrade cell had 3 priced events, too few to test.
Stage 2: the one survivor fails a date-matched control
The stage-1 placebo used the same stock at random dates. Auditor appointments cluster at AGM season, so it never controlled for when events happen. Stage 2 compares each event against the median of every other priced stock entered on the same session, with a sign-flip null:
| auditor change | n | median vs same-day peers | p | by year (120d) |
|---|---|---|---|---|
| 20 sessions | ~1,000 | −0.52% | 0.13 | 2022 −3.4 · 2023 −1.1 · 2024 −0.6 · 2025 −0.1 |
| 60 sessions | −0.84% | 0.14 | ||
| 120 sessions | −0.52% | 0.42 | 2022 +18.0 · 2023 +5.5 · 2024 +0.5 · 2025 −1.1 · 2026 −4.1 |
The −3.7-point gap was when the events happened, not what they said. Its sign flips across years. Independent-director resignations also show nothing: 120d −1.74%, p 0.46.
Two limits worth knowing
- NSE's announcement text is boilerplate. "…has informed the Exchange regarding Change in Auditors." Whether an auditor resigned mid-term (the real red flag) or rotated at an AGM is only in the attached PDF. These 2,066 raw events mix both, which dilutes any resignation effect. A PDF-reading follow-up could isolate mid-term resignations; the sample would then be small (tens, not thousands).
- Only about half the events price against the candle store (SME and BSE-only names drop out).
Product decision
- Show Reg-30 events (auditor change, director resignation, default, CIRP, rating action) on a symbol's page as dated facts with the filing link, the way trader-news shows 8-Ks. Do not call them warnings or signals: on 4.5 years of NSE data they predict nothing.
- Method lesson, adopted for every future event study: a same-stock random-date placebo is not enough. Use a same-date cross-sectional control. Here the weaker one manufactured a q = 0.018 "discovery".