The Bounce That Proves the Bear

A Composite Score of Nearly Zero

The Korean equity regime on 2026-03-10 scores $-0.08$ on the composite. The label is Transition. Five trading days ago, the label was also Transition. In between, the regime touched Mean Reversion. The market has traveled in a closed loop.

The 77.7% single-day advance ratio and a mean 1-day return of $+2.7$ across 2,774 symbols look constructive in isolation. They are not. The 5-day advance ratio is 20.3%. The 20-day advance ratio is 30.0%. The percentage of symbols above their 50-day moving average is 36.8% — nearly identical to the 37.1% reading five sessions ago. The breadth acceleration of $+6.32$ in the above-50-day-MA ratio over the last twenty-four hours reflects a single day's recovery inside a twenty-day deterioration. The Transition label is earned not by ambiguity but by arithmetic: the layer scores span from $-4.00$ to $+3.00$, and their sum is $-0.08$.

The Structure of the Transition

The daily path into this Transition is Mean Reversion — the prior session was classified as a mean-reversion environment before the composite recovered to near-zero. The weekly path is also Transition, meaning the intermediate-term regime has not meaningfully shifted. This is a structurally stable Transition: the market is not oscillating between two distinct states. It is oscillating within one.

On 2026-03-05, the composite was $+0.75$. Today it is $-0.08$. The five-session drift is $-0.83$. That drift has occurred despite a 77.7% advance ratio on the final day of that window. The implication is direct: each reflexive bounce in the Korean cross-section is recovering less composite regime score than the prior one. The recoveries are decaying. The Transition regime is not a neutral holding pattern — it is a regime under structural downward pressure that is absorbing upside sessions without confirming a directional reversal.

Layer-by-Layer Decomposition

The Trend layer scores Neutral at $+1.67$, an average across three index markets that are producing three distinct signals. KOSPI is Bull at 3/4 — its 60-day return is $+3.7$ and it sits above its 200-day moving average. KOSDAQ is Neutral at 2/4, but its 20-day return is $-7.6$: the index is holding a tenuous technical position against intermediate-term deterioration. KONEX is Bear at 0/4, with a 20-day return of $-4.5$. The Trend layer's $+1.67$ is the average of a Bull, a Neutral, and a Bear — three non-overlapping states compressed into a single composite number that, by construction, understates the divergence.

Breadth scores Mixed at $+1.00$. The 1-day advance ratio of 77.7% is the constructive input; the 36.8% above-50-day-MA figure is the constraining one. The breadth acceleration of $+6.32$ in the above-50-day-MA ratio is real. It does not yet change the aggregate level, which remains below any threshold associated with a healthy market structure. In the March 5 session, the 1-day advance ratio was 92.2% and the above-50-day-MA figure was 37.1%. Today, the advance ratio is 77.7% and the above-50-day-MA is 36.8%. The breadth readings are nearly identical to the bounce five sessions ago. The market has not progressed.

Volatility scores Stressed at $-4.00$. Proxy volatility is 80.87 on a 20-day basis, expanding to 124.84 on a 5-day basis. The short-term to long-term volatility ratio is 1.544 — classified as Short-term stress. The proxy volatility regime is independently classified as Crisis. The gap stress proxy is 54.54, meaning more than half the universe is experiencing overnight gaps that exceed typical intraday price ranges. Five days ago, proxy volatility was 79.13 with a short-term ratio of 1.495 and a gap stress proxy of 68.52. The volatility level has not declined — it has increased marginally on the 20-day measure and declined modestly on the gap proxy. The Crisis classification is stable.

Liquidity scores Thin at $-2.00$. The mean volume ratio is 0.70 — below the 0.92 reading from five sessions ago. The gap frequency is 54.5%. Low volume ratios combined with sustained gap frequency indicate that price discovery is occurring at the open, not through intraday order flow. Thin liquidity in a Crisis volatility regime produces a specific operational problem: the bid-ask spread at which a systematic fund can execute is wider during periods of elevated overnight gap risk, and the volume available to absorb institutional orders is below historical norms. The Liquidity score of $-2.00$ is the floor of the Thin classification.

Correlation/Dispersion scores Stock Picking at $+3.00$. Average pairwise correlation is 0.318. This is the one layer that has been consistently positive across every Transition session in the current cycle. It is also the layer that carries a structural caveat: with the top dollar-volume names capturing a disproportionate share of aggregate turnover, the headline correlation figure includes a large population of names that are too illiquid to trade at institutional size. The stock-picking signal is real. Its accessible version is narrower than 0.318 implies.

The S2 Problem

The KMeans clustering produces five state groups. The largest is S2: 1,247 symbols, predominantly KOSDAQ-listed, with a 5-day return of $-4.6$ and a 20-day return of $-13.3$. Only 9.9% of S2 members are above their 50-day moving average. S2 represents 45% of the 2,774-symbol universe. Nearly half of all traded Korean equities are technically broken on an intermediate-term basis, with fewer than one in ten above any medium-term support level.

S2 is the structural condition of the Korean cross-section. The single-day 77.7% advance ratio is a bounce inside S2, not an escape from it. A name that was $-13.3$ over twenty days and $-4.6$ over five days, advancing $+2$ in a single session, remains technically deteriorated. The numerics are additive: $-13.3$. The 20-day trend for the median S2 name has not reversed. It has become marginally less negative.

The S0 cluster — 912 symbols, KOSPI-dominant — has a 5-day return of $-5.4$ and a 20-day return of $-4.8$. 33.8% of S0 members are above their 50-day moving average. S0 and S2 together account for 2,159 symbols — 77.8% of the universe — and both are posting negative intermediate-term returns with below-average technical positioning. The visible Korean equity market, as measured by KOSPI large-caps, is in better shape than the broader universe because S0's 33.8% above-50-day-MA figure significantly exceeds S2's 9.9%. But neither cluster is constructive.

S3 and the Momentum Bifurcation

The counterpoint to S2 is S3: 88 symbols, KOSDAQ-dominant, with a 5-day return of $+21.7$ and a 20-day return of $+47.9$. Volatility in S3 is 123.5%. Volume ratio is elevated. Representative names include 우리기술, 현대ADM, and 성호전자.

S3 is 3.2% of the universe by symbol count. It is generating returns that are approximately five times the magnitude — with opposite sign — of the S2 cluster over the same twenty-day window. This is not market leadership. It is the statistical extreme of a bifurcated cross-section. In a healthy broad-market advance, 3% of the universe in a momentum cluster would represent the vanguard of a wider move. In a market where 45% of the universe is in the S2 state and the volatility regime is Crisis, S3 is a separate phenomenon operating by separate mechanics — most likely a convergence of small-cap names catching idiosyncratic catalysts in a thin liquidity environment.

The 20-day return of $+47.9$ with a 20-day volatility of 123.5% implies a Sharpe ratio of approximately $0.39$ on a 20-day annualized basis — positive but not compelling as a standalone allocation, particularly when the gap frequency environment means that realized Sharpe over intraday execution windows will underperform the return series computed at daily closes.

Energy Build-up: The Asymmetric Pressure

The emergency watchlist identifies 8 symbols with energy scores above 85 and explosion probability above 80. The directional composition of the top 40 energy accumulation names is: 25 downside-biased, 8 upside-biased, 7 neutral. The ratio is approximately 3-to-1 in favor of downside.

Energy build-up in this framework reflects compressed volatility states prior to a directional resolution — analogous to a coiled spring where the model measures the compression and assigns a probability to the decompression magnitude. The fact that 036690 carries energy 95.68 with explosion probability 96.33 and a downside bias means the model's estimate is that this name is highly likely to move, and the structural state is more consistent with a downward move than an upward one.

The five high-energy names on the watchlist span both directions — 036690 and 199290 downside, 278990 and 257990 and 245450 upside — but the directional skew of the broader top-40 cohort is not ambiguous. 62.5% of the high-energy names are downside-biased. In a market where the composite regime score is $-0.08$ and volatility is in Crisis, the energy model is consistent with the volatility layer: elevated realized volatility has not yet resolved the latent compression in a subset of names. When it does, the base case is downward.

Anomalous Price Action at the Extremes

The 5-day return extremes in the Korean cross-section are not marginal. 디에이치엑스컴퍼니(031860) is $+128.5$ over five sessions. 에이엔피(015260) is $+64.1$ with a volume ratio of 4.37 and an anomaly classification of 과열/급등형 — overheated momentum. At the opposite end, 세토피아(222810) is $-98.8$ over five days; it occupies its own cluster, S4, alongside 지더블유바이텍 at $-97.1$. 동성제약(002210) is $-44.9$ over five days with a single-day return of $-30.0$.

These readings are not useful as directional signals. They are useful as regime diagnostics. A cross-section where the top five-day return is $+128.5$ and the bottom is $-98.8$ is a cross-section in which mean-variance optimization assumptions are materially violated. Returns are not drawn from a stable distribution at this dispersion level. Position sizing frameworks that assume normally distributed returns will under-allocate to the extreme losers (correct) and over-allocate to the extreme winners (incorrect — the winners are anomalous, not persistent). The operational implication is to treat names with five-day returns beyond $\pm 50$ as regime outliers rather than signal inputs.

Signal Filter and ML Models

The signal filter top bullish names present a consistent technical picture. 대성미생물(036480) carries a decision score of 100 with 11 buy signals across MACD golden cross, RSI confirmation, Ichimoku breakout, price break above the 200-day moving average, Bollinger upper band breakout, ADX trend confirmation, CCI, Williams %R, Keltner channel breakout, RSI centerline bull, and positive momentum. This is a technically exhaustive signal stack — every momentum and trend-following indicator in the filter is simultaneously triggering.

When every indicator aligns at a regime score of $-0.08$ in a Crisis volatility environment, the interpretation requires caution. A broad technical alignment is most reliable when the macro environment supports directional follow-through. In a Transition regime with Thin liquidity and a gap frequency above 50%, the execution risk on high-signal names is elevated. The signal quality is not in question. The execution environment for acting on that quality is.

The RandomForest upside classifier achieves AUC of 0.539 and accuracy of 52.6% — modestly above chance, and marginally weaker than the 0.570 AUC and 55.1% accuracy from the March 5 session. The top features remain volume ratio (0.122), 5-day volatility (0.117), and volatility expansion (0.116). The model's edge has not disappeared, but it has compressed. The top upside candidates — 오건에코텍 at 64.1%, CJ CGV at 61.3%, 서희건설 at 61.0% — share the same profile identified five sessions ago: recent drawdown, moderate volume expansion, small-cap positioning. The model continues to identify oversold mean-reversion candidates at the micro-cap level.

The XGBoost return model assigns a mean expected 5-day return of $+0.3$ across the universe, with scenario offsets of $-8.6$ (bear) and $+3.9$ (bull). The bear scenario offset has widened from $-6.8$ on March 5 to $-8.6$ today — the model's conditional bear case has deteriorated. The bull scenario offset has narrowed from $+4.5$ to $+3.9$. The return model's expected value distribution is skewing more negatively than five sessions ago, consistent with the composite score drift of $-0.83$ over the same period.

Early Warning and Risk Pressure

The early warning module returns neutral: no distinct warning signals. Risk pressure has moved from $+5.94$ to $-1.42$ — a swing of $-7.36$ that reflects the single-day bounce. The breadth above-MA50 delta is $+6.32$, confirming the day's advance was real and broad enough to lift the technical participation rate meaningfully in a single session.

The neutral early warning classification, however, should not be read as an all-clear. The module is not detecting new stress accumulation. It is detecting the absence of acceleration in existing stress. The existing stress — Crisis volatility, 36.8% above-50-day-MA, downside-skewed energy build-up, S2 cluster at $-13.3$ over twenty days — remains. A reading of no new warning signals inside a market already in a Transition regime with these structural conditions means the damage is priced in and positioned, not that it has resolved.

Operational Implications

The trade plan records 18 buys and 0 sells, with 12 holds. All buys correspond to the bullish signal filter names. The absence of any sell signals in a market where the energy build-up skews 3-to-1 downside and the S2 cluster covers 45% of the universe reflects the signal filter's design: it identifies technically configured names, not regime conditions. The signal filter is functioning correctly. The regime environment should govern position size and gross exposure, not the directional count.

In practical terms: the Correlation/Dispersion layer at $+3.00$ continues to justify a long/short structure where idiosyncratic selection is the return driver. The Volatility layer at $-4.00$ in Crisis constrains gross exposure — position sizes derived from trailing 20-day volatility estimates are implicitly over-allocated at a short-term-to-long-term ratio of 1.544. The Liquidity layer at $-2.00$ constrains execution: mean volume ratio of 0.70 with a 54.5% gap frequency means that order routing at the open carries elevated slippage risk, particularly for names in the S3 momentum cluster where volume is abnormal. The 18-buy trade plan should be executed with reduced size relative to normal regime conditions, with gap limits on overnight exposure.

The bounce on 2026-03-10 is structurally identical to the bounce on 2026-03-05. The composite score is lower this time. The volatility level is higher. The ML return model's bear scenario has widened. The energy build-up is more heavily downside-biased. Five sessions of data separate these two nearly identical regime snapshots — and across that window, the underlying conditions have, on net, deteriorated.

A market that needs a 77.7% advance day to stay in Transition is not building toward confirmation. It is building toward a decision.