QUANT REPORT — THG (Thonburi Healthcare Group PCL)¶
Data window actually usable in this package: 60 sessions, 2026-03-26 → 2026-06-23 (55 real trading days after dropping 5 zero-volume carry-forward stubs flagged by audit). The 2,076-row 10y CSV is referenced at a local path but not provided in the package; no SET-index, sector, FX, oil, or rate series provided. Almost all "10y" statistics below are therefore not computed from the package — they are flagged as unavailable. Where I report a number, I show what data window produced it.
1. Return statistics¶
| Metric | THG (60d window, n=54 log-returns) | SET Index |
|---|---|---|
| Mean daily log-return | −0.20% | n/a — series not in package |
| Annualised drift (×252) | −50% (window-only, not representative) | n/a |
| Daily σ | ~2.5% | n/a |
| Annualised σ (×√252) | ~40% | n/a |
| Skew | negative (driven by 5/13 −4.5%, 6/19 −3.3%) — n≈54 too small for stable estimate | n/a |
| Kurtosis | elevated (one +9.6% day on 3/31 dominates) — unstable at n=54 | n/a |
| Sharpe (window) | strongly negative; not meaningful at this n | n/a |
| Window max DD | −21.9% peak-to-trough (9.35 on 3/31 → 7.30 on 6/23) | n/a |
| 10y vol / Sharpe / max DD | NOT IN PACKAGE — cannot report | n/a |
Caveat: this window is post-RAM-recap + post-Boon-overhang. It is a small, biased slice. Pre-2024 returns would be from a structurally different capital structure (110%+ more shares outstanding post-raises). Combining old and new into one 10y vol estimate is statistically questionable even if the file were present.
2. Volatility regimes¶
Cannot run an HMM or rolling-60d-vol segmentation — full 10y series not provided. From qualitative inflection points in the dossier + window:
| Regime | Approx. dates | Character |
|---|---|---|
| Pre-scandal | … → Sep 2024 | n/a in package |
| Crisis | Sep 2024 → Mar 2025 | Boon flees, arrest warrants, RAM PP announced |
| Recap / re-rating | Mar 2025 → late 2025 | PP at ฿8.65, RO at ฿5.00, share count +~111% |
| Post-recap fade (current) | ~Q1 2026 → now | window shows steady drift 9.35 → 7.30 on rising-volume down days; annualised σ ~40% |
Current regime: post-recap fade, elevated vol, negative drift.
3. Beta & factor exposures¶
Cannot compute. No SET, sector ETF (HELTH), USD/THB, oil, US10Y, or China property series in the package. The audit also confirms no peer multi-year price series, no income statement, no shareholder file. All of the following are therefore unavailable:
- β to SET (5y / 10y)
- β to oil, USD/THB, US10Y, China property
- Style factor ranks (value / size / momentum / quality / yield)
- Dividend yield rank
Qualitative prior only (not estimated from data): hospital names typically have β_SET in 0.6–0.9, low oil sensitivity, modest USD/THB sensitivity via medical-tourism revenue. THG-specific β is likely destabilised by the 2024–25 idiosyncratic shocks (scandal + recap) — regression betas across that window would be near-meaningless.
4. Mean reversion vs trend¶
- Hurst exponent / variance-ratio test: cannot compute from 55 returns with statistical reliability (need ≥ 500). Not reported.
- Window evidence: 12 of 14 down-days < −1% are followed by another down or flat day → short-term momentum, not mean reversion, in the current regime. But n is tiny.
- Structural anchors that any mean-reversion model would point to:
- RAM private-placement price ฿8.65 (Mar 2025) — informed-buyer print
- Rights-offering price ฿5.00 — discounted, dilution-driven
- 60-day VWAP-equivalent (simple mean of 55 closes) ≈ ฿7.94
- Dossier 52-wk high ฿38.50 is pre-recap and not a valid anchor for the post-dilution share base.
Implication: at the 2y horizon the relevant equilibrium is somewhere between the RO price (฿5.00) and the PP price (฿8.65), not the pre-dilution chart.
5. Drawdown profile¶
From the 60-session window only:
| Metric | Value |
|---|---|
| Window peak | ฿9.35 (2026-03-31) |
| Window trough | ฿7.25 intraday / ฿7.30 close (2026-06-23) |
| Window max DD | −21.9% |
| Days peak→trough | ~57 trading days, no recovery yet |
| Largest 1-day drop | −4.5% (2026-05-13) |
| Largest 1-day gain | +9.6% (2026-03-31) |
Full 10y max-DD: NOT IN PACKAGE. Given the dossier's ฿38.50 → ฿7.30 path, the 10y max DD is plausibly ~−80%, but I will not state that as computed.
6. 2-year forecast — model comparison¶
Assumptions stated for each model. Spot ฿7.30. σ used = 40% annualised (window estimate). r_f set to 0 for simplicity (Thai 10y not in package).
| Model | 12m point | 12m 90% CI | 24m point | 24m 90% CI |
|---|---|---|---|---|
| GBM, drift=0 (vol-only random walk, σ=40%) | ฿7.30 | [฿3.95, ฿13.50] | ฿7.30 | [฿3.10, ฿17.20] |
| GBM, window drift (μ=−50%/yr, σ=40%, naive extrapolation — do not trust) | ฿4.45 | [฿2.40, ฿8.20] | ฿2.70 | [฿1.15, ฿6.40] |
| Mean reversion to PP anchor (target ฿8.65, half-life ~12m, σ=40%) | ฿7.95 | [฿4.95, ฿12.75] | ฿8.40 | [฿4.45, ฿15.85] |
| Multi-factor regression | n/a — factor series not in package | — | — | — |
| Naive EPS × P/E (FY25 EPS ฿0.054 × peer 16x = ฿0.86; ×35x bull = ฿1.89; ×100x momentum = ฿5.40) | ฿0.86 – ฿5.40 | wide | depends on EPS trajectory not in package | — |
| Median of usable models | ~฿7.30–7.95 | [฿4, ฿13] | ~฿7.30–8.40 | [฿3, ฿16] |
Notes on each forecast¶
- GBM drift=0 is my preferred central case. It says: "spot is the best forecast; vol bands tell you the dispersion." 90% bands at 24m span roughly ฿3 → ฿17.
- GBM window-drift is shown for completeness but is structurally wrong: a 60-day window cannot pin long-run drift, and ฿2.70 sits below the rights-offering price RAM just paid — economically implausible without further dilution.
- Mean reversion to ฿8.65 assumes the RAM PP price is the informed-buyer fair value. This pulls 24m point estimate to ~฿8.40.
- Naive EPS×P/E demolishes the bull case at face value: at ฿0.054 EPS × peer 16x = ฿0.86. The market is already paying ~135x trailing — i.e. the recovery is priced in. To justify ฿7.30 needs net profit ~฿800m (8× FY25); to justify the owner's ฿30 needs ~฿3.3bn (35× FY25). EPS path not in package, so I cannot forecast it.
- GBM ignores fat tails. This stock has had idiosyncratic shocks (Boon flees; PP announced; recap completed) that drive returns far outside Gaussian. Real tails are wider than CIs shown — particularly to the downside (Boon legal verdict, RAM strategic shift) and upside (M&A premium if RAM moves to consolidate above 50%).
7. Where models disagree and why¶
- GBM-drift=0 (฿7.30) vs Naive EPS×P/E (฿0.86–฿5.40): market is pricing forward earnings recovery that is invisible in trailing fundamentals. If FY26 EPS doesn't double-or-better, the EPS×P/E model wins and price compresses toward ~฿2–฿5.
- Mean-reversion (฿8.40) vs GBM-window-drift (฿2.70): the PP anchor says "informed buyer paid ฿8.65 13 months ago" — a fundamentals-grounded floor. The drift extrapolation says "recent tape is bleeding" — a momentum signal. These are incompatible without resolution of the Boon legal overhang.
- All models assume no further dilution and no further share issuance. Share count has already grown ~111% in 18 months; another raise would compress all targets by ~the dilution factor.
8. Confidence: LOW¶
Limits of this analysis¶
- No 10y price file in this package — annualised return, vol, Sharpe, max-DD, Hurst, regime model are all uncomputable here. Numbers above are 60-session only.
- No benchmark / factor series — no β, no factor decomposition, no R².
- No financials beyond dossier line-items — EPS path, EBITDA, post-recap net debt all unverified; the Naive model uses one data point (FY25 EPS = ฿0.054).
- Structural break in 2025 (RAM PP + RO, share count +111%, debt repaid) makes any pre-2025 statistic mechanically incomparable to post-2025. A clean 10y vol/drift estimate is statistically not the right object for this name.
- Fat-tail risk both ways: Boon legal resolution, RAM strategic move (>50% consolidation or full TO), or a Thai hospital sector re-rating are all binary catalysts not captured by GBM.
- Stocktwits ticker collision (NYSE:THG = Hanover Insurance) — discarded entirely per audit.
Bottom line forecast (median of models I can compute, 24m): ฿7–฿9 central, 90% CI ~฿3–฿16. The owner's ฿30 sits >2 standard deviations above even the wide GBM band and requires an EPS trajectory not supportable from any number in this package.