Research & Writing
Ideas
Research notes. Market observations. Long-form thinking on quantitative finance, macro structure, and the mechanics of systematic investing.
Easing Pressure, Expanding Lows
US risk pressure collapsed 33 percentage points in five days while new 20-day lows remain above the warning threshold — and a 100% portfolio turnover has replaced treasuries with copper, nickel, and industrials.
Stock Picking at 4.25, With Thin Volume Underneath
The Korean composite regime score reached +4.25 on 2026-03-24 as correlation collapsed to 0.319 and dispersion widened — but 96.5% portfolio turnover and persistent thin liquidity complicate the clean read.
Walk-Forward Validation as Infrastructure
Walk-forward validation is not a backtest. It is a meta-test of backtests — and the criteria for failure are different from what most practitioners expect.
Alpha That Improves With Time
Not all signals decay. Some alphas strengthen with holding horizon — and knowing which ones is the foundation of a long-horizon systematic strategy.
Two Strategies Are Not Twice One Strategy
Mean reversion and accumulation-trend strategies exploit orthogonal market mechanisms. Running them simultaneously provides diversification at the strategy level, not just the portfolio level.
Volatility Targeting as the Primary Sharpe Lever
A 40.7% CAGR at Sharpe 0.62 is not a success. Diagnosing why reveals that volatility targeting is the dominant mechanism for risk-adjusted improvement.
Nonlinear Alpha Combination: Why the Risk Validator Said No
XGBoost beats linear IC-weighting on 4 of 5 OOS folds. The risk validator still rejected it. Here is why that outcome is correct.
Six Signals Across Four Assets
Replacing a 13-signal single-asset regime with 6 cross-asset indicators improves MDD from -21.9% to -20.0% while raising Sharpe from 1.04 to 1.16.
What Accumulation Signals Actually Predict
Rank IC and Pearson IC diverge sharply for accumulation-based signals. The gap reveals why IC-weighting fails and what the signals actually measure.
Alpha Factory v3: A Five-Gate Pipeline for Long-Only Signal Selection
How a template-based generation system producing 9,000 candidate alphas distills to 29 deployable signals — and why a liquidity archetype dominates the result.
The Decision Arrives, Conditionally
The Korean composite regime score jumped from -0.08 to +2.08 in a single session, shifting the label from Transition to Stock Picking — but Crisis volatility and Thin liquidity have not moved.
The Pharma Rotation Resets Itself Overnight
On 2026-03-11, US breadth sits at 34% above MA50 with a 4:1 new-low-to-new-high ratio — but the bullish signal cohort has completely replaced itself since March 10, and bond ETFs have disappeared from the portfolio entirely.
The Bounce That Proves the Bear
On 2026-03-10, the Korean market posted a 77.7% advance ratio and a mean 1-day return of +2.7%. The composite regime score is -0.08. The energy build-up skews 25-to-8 toward the downside.
New Lows at Five Times the Rate of New Highs
On 2026-03-10, 24.37% of the US equity universe printed new 20-day lows against 5.18% new highs. The signal filter finds bullish pockets — but they are telling a defensive story.
Machine Alpha and the Limits of Learned Signals
AI has entered quantitative investing at every layer. The evidence on whether it generates durable alpha is more complicated than the capital flows suggest.
The 92% Advance Ratio That Changes Nothing
On 2026-03-05, the Korean market posted a 92.2% single-day advance ratio. The composite regime score is 0.75. These two facts are not in conflict — they are the same warning.
Stock Picking When the Best Picks Are Cash
The regime reads Stock Picking, but the ML upside model's top candidates are T-bill ETFs — a precise signal about where the cross-sectional opportunity actually sits.
Variance Drain and the Rebalancing Premium
Geometric returns are not arithmetic returns. The gap is exploitable.
When the Regime Disagrees With Itself
Transition regimes are not regime-free. They carry their own structure — and their own penalties for misreading them.