The 92% Advance Ratio That Changes Nothing

A Number That Looks Definitive

On 2026-03-05, 92.2% of the Korean equity universe advanced in a single session. The mean 1-day return across 2,773 symbols was $+7.5$. The median was $+7.3$. On the surface, this is a broad, symmetrical, high-conviction advance — the kind of breadth reading that typically precedes regime confirmation.

The composite regime score is $+0.75$. The regime label is Transition.

These are not contradictory. They are the same statement about the same market.

The 5-day advance ratio is 9.7%. Over the last five trading sessions, fewer than one in ten Korean equities has advanced. The 20-day advance ratio is 35.9%. The percentage of symbols above their 50-day moving average is 37.1% — down sharply from any constructive baseline. New 20-day lows (3.6%) marginally outnumber new 20-day highs (2.6%). The single-day 92.2% advance ratio is the inversion of a 5-day reading of 9.7%. It is a reflexive bounce, not a regime shift. The Transition label is correct. The 92% print is the mechanism that generates the label, not the evidence against it.

Two Prior Regimes, One Destination

The current regime is Transition. What makes this particular Transition structurally distinct is the path that produced it.

On a daily basis, the previous session was classified as Mean Reversion. The composite score moved from $-2.42$ to $+0.75$ — a swing of $+3.17$, driven primarily by Breadth recovering from Weak to Mixed ($+4.00$ contribution) and offset partially by Liquidity deteriorating from Normal to Thin ($-1.00$) and Volatility worsening within Stressed ($-1.00$). On a weekly basis, the prior regime was Stock Picking — itself a snapshot that captured a high-dispersion, low-correlation environment from a week ago. The composite score on that weekly lookback has deteriorated by $-5.00$, with Liquidity falling from Liquid to Thin ($-2.00$), Volatility from Stressed to a worse Stressed reading ($-2.00$), and both Trend and Breadth weakening.

Two separate prior regimes — Mean Reversion and Stock Picking — have converged to the same current label: Transition. This matters for interpretation. A Transition regime that arrives from a single prior state represents directional uncertainty. A Transition that arrives from two distinct prior states, measured at different frequencies, represents a market in genuine structural reorganization. The daily path says: the selling impulse exhausted itself into a single-day snapback. The weekly path says: the macro and liquidity conditions that supported Stock Picking four weeks ago have materially deteriorated.

The bounce on 2026-03-05 did not resolve either path. It interrupted both.

The Layer Scores and What They Measure

The composite score of $+0.75$ is the sum of five layer readings that point in different directions.

Trend scores Neutral at $+2.00$. KOSPI is Bull-classified with a score of 4/4 — its 60-day return is $+6.6$, its 20-day slope is $+7.9$, and it sits above both its 50-day and 200-day moving averages. KOSDAQ is Neutral at 2/4, with a 20-day return of $-6.5$ and a 20-day slope that remains positive at $+4.4$ despite an intermediate-term breakdown. KONEX is Bear at 0/4: negative at every trend dimension. The Trend layer is not neutral because signals are ambiguous — it is neutral because three index markets are producing three distinct trend states simultaneously.

Breadth scores Mixed at $+1.00$. The single-day advance ratio of 92.2% is the top input, but the layer also incorporates the above-50-day-MA figure of 37.1% — which remains deeply below any constructive threshold. These two sub-signals are in direct opposition, and the layer's Mixed score reflects exactly that.

Volatility scores Stressed at $-4.00$ — the single largest negative contributor. Proxy volatility sits at 79.13 on a 20-day basis, expanding to 118.32 on a 5-day basis. The short-term to long-term volatility ratio is 1.495, classified as Short-term stress. The proxy volatility regime is independently classified as Crisis. The gap stress proxy — derived from overnight gap frequency — is 68.52. These are not marginally elevated figures. The volatility regime is in a structural expansion cycle, and the single-day bounce did not alter any of these inputs.

Liquidity scores Thin at $-1.00$. The mean volume ratio across 2,773 symbols is 0.92, the median is 0.77. Only 9.8% of the universe traded at 1.5 times its 20-day average volume — unusually low participation at the individual name level, even as aggregate dollar volume ($46.2$ trillion KRW 20-day average) appears headline-adequate. The high-volume concentration is driven by a small number of anomalous names. Aggregate volume figures in a thin liquidity regime are misleading in the direction of overstating tradeable capacity.

Correlation/Dispersion scores Stock Picking at $+3.00$. Average pairwise correlation is 0.306, median is 0.322. Cross-sectional dispersion is 6.2% on a 1-day basis, rising to 9.8% over five days and 22.4% over twenty days. A correlation of 0.306 in the Korean market is classified as Mixed sector regime — consistent with a market where sector-level co-movement is present but not dominant. At this dispersion level, idiosyncratic variance contributes meaningfully to total return variance at the individual position level.

The composite of $+0.75$ is nearly zero. The layer scores span from $-4.00$ to $+3.00$. This is not a market that is uniformly Transitioning — it is a market where Transition is the only honest label for a set of signals that cannot be reconciled into a single directional read.

The Defense-Energy Cluster

The KMeans state clustering on three PCA components (explained variance: 19.9%, 13.9%, 11.8%) produces five groups. Cluster S4 is the structural story of this session.

S4 contains 113 symbols, predominantly KOSDAQ-listed, with a 5-day return of $+18.1$ and a 20-day return of $+40.9$. Twenty-day volatility is 104.0% — high even relative to the already-stressed Korean cross-section. Volume ratio is 2.26. 98.2% of S4 members are above their 50-day moving average. Representative names are 한화시스템(272210), 흥구석유(024060), and LIG넥스원(079550).

This cluster is the only technically unbroken momentum cluster in today's cross-section. Its composition tells the story: Korean defense contractors and energy names. 한화시스템 gained $+30.0$ on the day with a 5-day return of $+35.4$. LIG넥스원 gained $+23.3$ on the day with a 5-day return of $+49.9$. The 5-day strongest symbols list confirms the energy component: 한국석유 at $+61.3$, 중앙에너비스 at $+56.2$, 흥구석유 at $+55.0$, 대성에너지 at $+51.0$.

These are theme-driven, idiosyncratic moves concentrated in two adjacent sectors — defense and domestic energy. They are not broad market leadership. They are the visible portion of a sector rotation that is occurring inside an otherwise stressed cross-section. The S4 cluster's $+18.1$ 5-day return is not a market phenomenon. It is a factor phenomenon: elevated geopolitical risk premium being re-priced into Korean defense and energy assets.

The practical consequence for a long/short book is asymmetric. S4 names have already moved. The momentum signal is present but the volatility-adjusted entry window has narrowed. S4 at 104.0% 20-day volatility, combined with a gap frequency regime (68.5% of the universe gapping more than 2% at the open) means that adding to these names at the current price requires an explicit assumption about the continuation of the thematic catalyst — not the market regime. A regime-based position sizing framework does not provide that input.

The other clusters tell the opposite story. S3 contains 746 symbols, KOSPI-dominant, with a 5-day return of $-9.9$ and only 32.3% above their 50-day moving average. S0 contains 1,084 symbols with a 5-day return of $-11.4$ and only 6.3% above their 50-day moving average. Between S0 and S3, more than 1,800 symbols — 65% of the universe — are technically deteriorating on an intermediate-term basis. S4's visibility is high precisely because the rest of the market is not participating.

Volatility Regime: The Crisis Classification

The proxy volatility regime classification is Crisis. This requires context. The Korean proxy volatility construct — built from cross-sectional realized volatility rather than a direct VIX equivalent — reads 79.13 on a 20-day horizon, expanding to 118.32 on a 5-day basis. The mean 20-day volatility across individual symbols is 79.1%, with a median of 72.5%. Mean volatility expansion (5-day vs. 20-day) is 1.53; the median is 1.64.

A volatility expansion ratio of 1.53 at the mean means that the average Korean equity has realized 53% more volatility over the last five trading days than its 20-day trailing average. That is not a spike. It is a sustained acceleration. The gap stress proxy at 68.52 confirms this: 68.5% of the universe opened more than 2% away from the prior close. More than two-thirds of tradeable names are experiencing overnight price discovery that exceeds their typical intraday range.

For strategies that size positions based on volatility estimates, the practical problem is which volatility estimate to use. The 20-day estimate understates current realized risk by 53%. The 5-day estimate captures the current stress regime but will overstate risk when (if) the regime normalizes. The standard approach — exponentially weighted moving average with a decay factor calibrated to half-life — produces an intermediate estimate, but in a regime where volatility is both elevated and accelerating, the EWMA lag remains structurally problematic. Position sizes derived from any trailing volatility measure are, in the current environment, implicitly over-allocated relative to the forward risk surface.

The operational implication is direct: volatility-targeting portfolios should reduce gross exposure at current vol levels not as a precautionary adjustment but as a mechanical output of their own sizing rules, if those rules are correctly specified.

Correlation Structure in the Korean Cross-Section

Average pairwise correlation of 0.306 in the Korean market occupies a structurally different position than 0.306 would in a US equity context. The Korean universe contains 2,773 symbols across KOSPI, KOSDAQ, and KONEX — a cross-section that includes large-cap conglomerates, mid-cap industrial names, and micro-cap KONEX issuers with median dollar volume of approximately 1.0 billion KRW per day. Across this heterogeneous universe, a correlation of 0.306 reflects the natural structural correlation of a market with dense chaebol ownership linkages, shared macro sensitivity, and a concentrated institutional investor base.

The cross-sectional standard deviation over 1 day of 6.2% — rising to 9.8% over 5 days and 22.4% over 20 days — reflects genuine dispersion, but the distribution is asymmetric. The top 1.4% of symbols by positive return captured 1.4% of aggregate dollar volume. The top 10 dollar-volume names captured 40.1% of aggregate dollar volume. The gap between the dispersion in returns and the dispersion in dollar volume is wide. High cross-sectional return dispersion with heavily concentrated dollar volume means that the dispersion signal is dominated by names that are too small to trade at scale.

This is the structural constraint on Correlation/Dispersion-based opportunity in the Korean market. A $+3.00$ Correlation/Dispersion score implies favorable stock-picking conditions. In an environment where 40% of aggregate dollar volume is concentrated in the top 10 names by daily turnover, the effective stock-picking opportunity available to a systematic fund with meaningful AUM is narrower than the headline dispersion figure suggests. The 6.2% cross-sectional standard deviation includes KONEX names with median volume of 1.0 billion KRW — well below the threshold for institutional size.

ML Signal Architecture

The RandomForest upside classifier achieves validation AUC of 0.570 and accuracy of 55.1% — modestly above chance but consistent with positive edge on a cross-sectional basis. The top feature importances are volume ratio (0.124), 5-day volatility (0.122), and volatility expansion (0.111). The model is sorting primarily on recent volatility dynamics and liquidity signals, not price momentum.

The top upside candidates share a structural profile: recent drawdown ($-5$ to $-20$ over five days), moderate volume expansion, and small-cap or micro-cap listing. 오건에코텍(212310) at 63.9% upside probability with a $-20.5$ 5-day return; 서희건설(035890) at 60.9% with a $-8.0$ 5-day return and a volume ratio of 5.40; 정원엔시스(045510) at 59.1% with a $-14.2$ 5-day return and a volume ratio of 6.31. The model is identifying oversold small-cap names with volume spikes — a mean-reversion signal operating at the micro-cap level, not a momentum signal.

The XGBoost return model assigns a mean expected 5-day return of $+0.9$ across the universe, with scenario offsets of $-6.8$ (bear) and $+4.5$ (bull). The top feature importances are new 20-day high (0.437) and market indicator variables (KOSDAQ: 0.341, KONEX: 0.321, KOSPI: 0.293). The return model's dominant feature is the 20-day high flag — a technical breakout signal. Given that only 2.6% of symbols are currently printing new 20-day highs, the model's upside is heavily concentrated in the S4 cluster names that are already in strong momentum states.

Together, the upside probability model and the return model are identifying two separate opportunity profiles: beaten-down micro-cap mean reversion (upside model) and defense/energy momentum continuation (return model). These are not the same trade. They require different position structures, different holding periods, and different risk limits.

Actionable Structure

The composite score of $+0.75$ is close enough to the Transition boundary that any single-day input shift could reclassify the regime. That instability is itself an input.

The one layer that has been consistently positive across both daily and weekly lookbacks is Correlation/Dispersion. The stock-picking signal has not degraded — it has remained at $+3.00$ through a regime that shifted from Stock Picking to Transition on a weekly basis. The other four layers have all deteriorated in at least one lookback window. The structural implication is the same as it was in the prior Transition session: Correlation/Dispersion provides the directional opportunity; Volatility and Liquidity impose the capacity and execution constraints.

In practice, this means: maintain hedged long/short structure where idiosyncratic selection is the return driver; treat the defense-energy theme (S4) as a concentrated sector position requiring explicit thematic conviction, not a regime-driven allocation; apply volatility-adjusted position sizing that accounts for the current 1.53 expansion ratio rather than trailing 20-day estimates; and maintain strict overnight gap limits given the 68.5% gap frequency environment.

The 92.2% advance ratio is real. It is the market's reflexive response to five days of 9.7% cumulative advance breadth. It does not resolve the Volatility layer. It does not restore Liquidity. It does not change the fact that 65% of the universe is below its 50-day moving average.

A single-day read of 92% tells you that sellers rested. It does not tell you that the regime changed.