The Only Thai Signal That Survived Regime Splitting

⚠️ Personal research and trading journal — not investment advice. The author does not provide licensed advisory services.

When I first started testing signals on Thai stock data, I made the same mistake most quantitative researchers make: I tested signals on the full sample, found something that worked, and declared it real.

The problem with that approach is that markets have regimes. A signal that looks positive in aggregate can be strongly positive in bull markets and strongly negative in bear markets — with the two effects roughly canceling, leaving you with a "mildly positive" result that disguises a regime-dependent strategy you haven't understood.

This is why I started splitting every finding by market regime before declaring anything real.

After splitting 30+ candidate signals on Thai data, almost all of them degraded or disappeared. One survived.

What Regime Splitting Reveals

The setup: I define two regimes for the SET market.

Uptrend (regime UP): SET index is above its 50-day moving average AND the 50d is above the 200d. Both conditions simultaneously — the index is trending up on the short-term, and the longer-term slope is rising.

Correction (regime DOWN): Anything else. SET below 50d, or 50d below 200d, or both.

For every signal I test, I measure: - Performance in regime UP - Performance in regime DOWN - Pooled average

The pooled average is almost meaningless. What matters is whether the signal performs consistently within each regime — and crucially, whether it reverses or simply degrades in the other regime.

If a signal is genuinely regime-dependent, it should work in one regime and either flatten or reverse in the other. Flat would mean "this signal is only useful when the market cooperates." Reversal would mean the opposite.

The Signal: Character Change

A "character change" describes a stock that was acting like the broad market (distributing, weak, consolidating) and suddenly begins to act differently. Specifically: after a period of relative weakness, the stock starts making moves that outpace the index.

The exact definition I used: the stock's 20-day RS score has been below 50 (acting weak) and now crosses above 50 (re-emerging). This isn't a new high in RS — it's a re-emergence from weakness. The stock was broken and is starting to heal.

On the pooled Thai data sample, character change showed modest positive performance. Not exciting — but not false either.

When I split by regime:

In uptrend conditions: the signal is real. The confidence interval excludes zero — +1.25% on the low end, +4.21% on the high end. This was after bootstrap resampling, not a single-sample number.

In correction conditions: the signal reverses. The median goes negative. The CI now spans zero (possibly noise in either direction) but the central estimate is −1.48%.

This is exactly what a legitimate regime-dependent signal looks like: positive in one regime, reversal in the other, with the pooled average being a misleading blend of the two.

Why Pooled Signals Disappear

I tested signals that looked promising in aggregate: RS momentum, volume surge following a quiet period, 21-day pullback to a rising average, high-RS stocks near pivot, specific chart patterns.

Most of them showed no regime dependence at all — they worked roughly the same (or not at all) across both regimes. That means whatever they were capturing wasn't market-state sensitive. It was either noise, or a real signal that's so persistent it doesn't care about regime.

The character change signal behaved differently. In regime UP, it was capturing something about leadership emerging from a recovering stock universe. When the market is in uptrend, a stock that was weak and starts to show re-emerging RS is identifying itself as a potential new leader. That's a real market dynamic.

In regime DOWN, the same signal means something different. A stock crossing RS 50 during a correction is often bouncing in a bear-market rally. It's participating in temporary recovery, not emerging as a new leader. The signal reverses because the underlying dynamic is different.

This is why regime is the most important variable to control for before declaring any signal real.

What This Changes in Practice

For Thai stock scanning, the character_change signal is now used only in regime UP (both conditions: SET above 50d AND 50d above 200d). In regime DOWN, the signal is ignored — or treated as evidence that any bounce is suspect.

The signal serves as a candidate-seed filter: it widens the net to include stocks just below the RS 80 threshold that are showing re-emergence. These are the stocks that may cross RS 80 in the next few weeks and become primary candidates.

It is not a direct entry signal. The full process still applies: chart check, volume confirmation, stop location, position sizing. But character change in regime UP is now a legitimate reason to put a stock on the watch list.

The Broader Lesson

Regime splitting is uncomfortable for quantitative researchers because it reduces sample size and forces you to admit that most signals are conditional.

But that's the point. Trading is conditional. Every edge I've found on Thai data has a market-state dependence. Pretending otherwise is how you convince yourself a strategy works when it only works half the time.

The character change × regime UP combination is the first Thai signal that survived this level of scrutiny: positive CI in the favorable regime, reversal in the unfavorable one, tested on multiple bootstrap samples. That's the standard everything else gets held to.

So far, most things fail it.

Track. Study. Wait. Strike.

Personal research and trading journal — not investment advice. The author does not provide licensed advisory services. — MOEasymmetry

Draft 2026-06-12. Signal: RS score crossing from <50 to >50 on Thai stocks. Regime defined as SET index vs 50dMA AND 50dMA vs 200dMA. Bootstrap CI from bootstrap resampling (n > 200 regime-UP observations). Candidate-seed use only — not a standalone entry signal.