The Decision Arrives, Conditionally

A Jump of 2.17 in One Session

The Korean equity regime on 2026-03-11 scores $+2.08$ on the composite. Twenty-four hours ago it scored $-0.08$. The label has shifted from Transition to Stock Picking. The magnitude of the single-session move — $+2.17$ — is large enough to warrant direct examination of what changed, what did not, and whether the change is structural or another decay-masked bounce inside the existing cycle.

The short answer is that one layer moved decisively, a second layer improved marginally, and two layers did not move at all. The composite shift is real. Its durability depends entirely on which of those four layers leads from here.

What Changed: Breadth

On 2026-03-10, the 5-day advance ratio was $20.3$. On 2026-03-11, it is $88.3$. That is not a marginal improvement. The 1-day advance ratio is $67.6$. The breadth layer has moved from Weak to Mixed, contributing a score of $+1.00$.

The breadth-above-MA50 delta is $+10.45$ over the last twenty-four hours — the largest single-session acceleration in this metric recorded during the current cycle. Two sessions ago that reading was $+6.32$; five sessions ago it was similar. Each prior acceleration resolved downward within a session or two. The difference this time is that the 5-day advance ratio has crossed $88$, a threshold associated with genuine cross-sectional participation rather than single-session relief. On March 5, the advance ratio reached $92.2$ on a 1-day basis but the 5-day figure remained depressed. Today the 5-day figure is the constructive one. That distinction matters for assessing whether the breadth has found a floor or is executing another reflex.

The Correlation/Dispersion layer holds at Stock Picking with a score of $+3.00$. Average pairwise correlation is $0.318$, unchanged in character from prior sessions. One-day cross-sectional standard deviation is $4.7$. The dispersion environment continues to reward stock-level selection over index-level positioning. This layer has been consistently the strongest positive contributor across every session in the current cycle, and its persistence is notable: it implies the cross-section is not moving as a single correlated block, which is a necessary condition for an active selection strategy to generate edge.

The Trend layer has strengthened. KOSPI is now classified Bull at $4/4$ — its 60-day return is $+5.5$ and it sits above its 200-day moving average. KOSDAQ is Bull at $3/4$ with a 20-day return of $-4.4$, improving from its $-7.6$ reading on March 10 but still negative on the intermediate horizon. KONEX remains Bear at $0/4$. The Trend layer's composite reads positive for the first time in the current cycle on the back of KOSPI's full Bull classification and KOSDAQ's improvement.

What Did Not Change: Volatility and Liquidity

Proxy volatility is $81.11$ on a 20-day basis. On 2026-03-10, it was $80.87$. The change is $+0.24$ — statistically negligible. The volatility layer scores Stressed at $-3.00$, and the independent volatility regime classification remains Crisis. Liquidity scores Thin at $-1.00$, reflecting a volume ratio below historical norms and a gap frequency that continues to dominate price discovery at the open rather than through intraday order flow.

These two layers are not lagging the regime shift. They are simply not participating in it. A composite score of $+2.08$ built on Breadth Mixed, Dispersion Stock Picking, and improved Trend — while Volatility remains at $-3.00$ and Liquidity at $-1.00$ — is a Stock Picking regime with load-bearing structural constraints. Position sizing frameworks that use trailing volatility to calibrate exposure are operating with an implicit haircut that the composite number does not communicate directly.

The practical arithmetic: in a Stock Picking regime with normal volatility and liquidity conditions, a systematic fund might target gross exposure consistent with the regime's historical Sharpe profile. In the current configuration, the volatility is approximately $81$ on a proxy basis while the correlation is $0.318$. The portfolio-level volatility for an equally-weighted long/short structure at unit gross is not dramatically lower than single-name volatility — the benefit of diversification is partially cancelled by the elevated cross-sectional volatility baseline. The correct response is to reduce per-position size relative to a normal Stock Picking environment, not to apply full regime-standard allocations.

The Cluster Map

The ML state clustering on 2026-03-11 presents a substantially different picture than March 10. The largest cluster is now S1: $1,191$ symbols, predominantly KOSDAQ-listed, with a 5-day return of $+6.8$ and a 20-day return of $-11.7$. S1 is the broken middle — names that have bounced on the 5-day horizon but remain structurally impaired on the 20-day. The 20-day return of $-11.7$ positions S1 as the direct successor to the S2 cluster that dominated the March 10 picture at $-13.3$ over twenty days. The sign is the same; the magnitude has compressed by $1.6$ percentage points. This is partial repair, not recovery.

S0 — $865$ KOSPI-dominant symbols — carries a 5-day return of $+5.1$ and a 20-day return of $-3.2$. The KOSPI large-cap tier is now posting a 20-day return that is negative but approaching flat. S0 is the most constructive cluster in the universe for a systematic long with a multi-week horizon. Blue-chip stabilization at $-3.2$ over twenty days, in a $+81$ proxy volatility environment, is not a strong absolute signal, but it is directionally different from what every prior session in this cycle produced.

S4 is the momentum cluster: $604$ symbols, KOSDAQ-dominant, with a 5-day return of $+15.3$ and a 20-day return of $+13.0$. This cluster represents the upper tail of the current bounce. The $+13.0$ 20-day return is a genuine intermediate-term positive reading — not a single-session anomaly. S4 names are the structural beneficiaries of the regime shift, and they account for $604$ of the $2,775$ symbols tracked — approximately $21.8$ of the universe.

S2 has collapsed to $113$ symbols with a 5-day return of $+42.0$. This is the extreme momentum tail: a small cohort executing parabolic moves inside a still-volatile market. S2 at $113$ symbols versus the prior session's dominant large cluster is a structural reorganization of the cross-section. Names that were distributed across the distressed cluster have either migrated toward S1 recovery or to S4 continuation. S3 is $2$ symbols at approximately $-98$ over twenty days — the blowup residual. This population has stabilized in count, suggesting the capitulation wave has not yet produced new extreme losers in volume.

Signal Filter: Bullish Skew at the Micro-Cap Level

The signal filter on 2026-03-11 shows a clear bullish skew with minimal sell-side activity. 이루온 (065440) carries a decision score of $100$ with $8$ simultaneous buy signals. 디케이앤디 (263020) scores $100$ with $7$ buy signals. 기산텔레콤, 코위버, and 에프알텍 each register decision scores of $100$. The dominant signals firing are MACD golden cross (signal 104), Ichimoku cloud breakout (189), CCI buy (411), and Williams %R buy (281) — a combination of trend-initiation and momentum-confirmation signals that is consistent with the S4 cluster's structural profile.

Sell signals are sparse: only $19$ MACD dead cross events and $52$ Ichimoku breakdown events appear across the universe. The signal count asymmetry — heavy on initiating buy signals, negligible on confirming sell signals — is a direct technical reflection of the regime label. In a Stock Picking environment with $88.3$ five-day advance participation, the signal filter is doing what it is designed to do: it is identifying the names where technical momentum has organized sufficiently to create an actionable asymmetry.

The trade plan registers $11$ buys and $0$ sells. Top executions include 디케이앤디 (263020), KB제29호스팩 (478390), 파인엠텍 (441270), 우리넷 (115440), and 센서뷰 (321370). The common profile is recent momentum confirmation with KOSDAQ positioning — names that have organized into S4-consistent structure with signal confirmation. In normal regime conditions, an $11/0$ buy-to-sell ratio on a $+2.08$ composite day is appropriate. In this environment, position size should be reduced by a factor consistent with the ratio of current proxy volatility to historical Stock Picking regime volatility.

ML Models: Modest Edge, Asymmetric Scenarios

The RandomForest upside classifier achieves AUC of $0.571$ and accuracy of $54.4$. These figures are marginally better than the $0.539$ AUC from March 10 — a recovery toward the $0.570$ level seen on March 5. The top features driving the model are volume ratio ($0.122$), volatility expansion ($0.116$), and 5-day volatility. The feature importance structure has not changed: the model continues to find its edge in volume-normalized momentum under conditions of expanding volatility — precisely the mechanics operating in the S4 cluster.

The XGBoost return model assigns a mean expected 5-day return of $+0.5$. The bull scenario offset is $+4.7$ and the bear scenario offset is $-8.4$. The bull scenario has expanded from the $+3.9$ recorded on March 10; the bear scenario has also widened from $-8.6$ to $-8.4$, a marginal improvement. The expected return distribution is slightly more positively skewed than the prior session, consistent with the regime shift. However, the asymmetry remains pronounced: the conditional bear scenario is approximately $1.8\times$ the magnitude of the conditional bull scenario in absolute terms. The model does not expect a symmetric recovery even in the new regime.

Energy Build-up: Downside Skew Persists

The emergency watchlist identifies $7$ symbols with energy scores above $85$ and explosion probability above $80$. The directional composition of the top $40$ high-energy names is: $23$ downside-biased, $11$ upside-biased, $6$ neutral. The ratio is approximately $2$-to-$1$ in favor of downside.

This is a meaningful change from the $3$-to-$1$ downside ratio on March 10, where the top $40$ showed $25$ downside and $8$ upside. The improvement is directionally consistent with the regime shift. It is not yet symmetric. A Stock Picking regime with $2$-to-$1$ downside energy skew is a regime where the latent decompression risk in the cross-section has not yet resolved. Energy build-up in this framework measures the compression of volatility prior to a directional release. The compression is easing, not resolved.

The early warning module returns Neutral. Risk pressure has eased by $-8.13$ percentage points — from $+5.94$ to $-2.19$. This is the largest risk-pressure easing in the current cycle's recorded data and constitutes the clearest quantitative signal that the regime label is not a false positive. A $-8.13$ swing in risk pressure, combined with a breadth-above-MA50 delta of $+10.45$ and a composite score jump of $+2.17$, describes a session that is genuinely different from the March 5 and March 10 bounces.

The Price Extremes as Regime Diagnostics

The return distribution at the tails remains non-Gaussian. 디에이치엑스컴퍼니 (031860) is $+202.6$ over five sessions. 센서뷰 (321370) is $+94.7$. 대한광통신 (010170) is $+88.4$. At the negative extreme, 세토피아 (222810) is $-99.1$ over five days and 지더블유바이텍 (036180) is $-97.6$. The cross-sectional return standard deviation of $4.7$ for a single session, in the presence of these tail observations, implies a distribution with substantially heavier tails than a normal approximation would produce.

The operational interpretation has not changed from prior sessions: names with five-day returns exceeding $\pm 50$ are regime outliers, not signal inputs. They are informative as diagnostics — their presence indicates the cross-section is operating in an environment of elevated idiosyncratic risk — but their inclusion in portfolio construction frameworks assumes a return distribution that does not exist at current dispersion levels.

A Qualified Transition

The Korean market has, for the first time in the current cycle, shifted into a definitively positive regime state. The composite score of $+2.08$ is not ambiguous. The $88.3$ five-day advance ratio is the most constructive breadth reading of the cycle. The S4 momentum cluster has grown to $21.8$ of the universe with a positive 20-day return. The risk pressure easing of $-8.13$ percentage points is the sharpest single-session improvement recorded in this cycle. These are not recycled observations from prior bounces. They are structurally new.

But the volatility layer has not moved. Proxy vol at $81.11$ in Crisis classification, combined with Thin liquidity, means the execution environment for acting on a Stock Picking signal is not the execution environment that a Stock Picking composite score of $+2.08$ implies in normal conditions. The energy build-up skews $2$-to-$1$ downside. The XGBoost bear scenario is $-8.4$ against a bull scenario of $+4.7$.

The regime has reached its decision. Whether the decision holds depends on whether volatility follows breadth lower, or breadth reverts toward volatility.