QAML Model: A Cross-Market Stock Selection Framework Combining Quant Factors, Analyst Revisions, and Machine Learning

Key Conclusions

The QAML model integrates academic factor signals, analyst earnings revisions, and AdaBoost machine learning to construct an industry- and beta-neutral long-short strategy, generating consistent excess returns across U.S., European, and Japanese equity markets. The core innovation lies not in the performance of any single factor, but in the three-tier architecture (core indicators, dynamic indicator pool, machine learning) that non-linearly adapts to shifting factor environments. Test data for 2025-2026 shows that the European Operating CF Yield strategy delivered annualized returns of +24% (2025) and +17% (2026 through March), while analyst revision signals have risen to the top tier in factor ranking systems for both the U.S. and Japan. The key investment implication of this framework is that static factor allocation is being replaced by dynamic environment-aware allocation, and quant investors should treat macro scenario dependencies as core decision variables for factor weights.

What the Market May Be Underestimating

The market's linear combination mindset for traditional factor strategies (e.g., equal-weighted value + momentum + quality) fails to capture two critical realities: First, the return asymmetry of academic factors under different macro states (direction of interest rates, inflation trends, economic cycle phases) is extreme. Second, the non-linear interaction between analyst revision signals and quant factors through machine learning releases far more alpha than simple summing. The current factor ranking system reveals that the U.S. EPS Revisions factor jumped from rank 10 to rank 1 (out of 58 factors) within one month, while Japan's 12M Fwd Earnings Yield rose to rank 1, indicating the market is still pricing the persistence of these signals rather than their structural strength.

Logical Chain: Factor Environment Dependency and Model Adaptability

Factor returns show statistically extreme asymmetry across macro states

Traditional factor returns diverge dramatically when interest rates, inflation, and economic cycles shift. The U.S. Book/Price value factor yielded an annualized return of 9.2% (t-stat 1.7) during rising long-end rate periods and -7.3% (t-stat -1.6) during falling rate periods—a spread exceeding 16 percentage points. The global Composite Value factor delivered 20.4% (t-stat 2.7) in recovery phases and -1.5% (t-stat -1.9) in recession phases—a gap of nearly 22 percentage points. Japan's low-volatility factor had a full-sample long-term average return of -3.3%: only 2.6% during rising inflation and -8.5% during falling inflation—demonstrating the factor's inherent failure in this market, where any static allocation would systematically underperform.

Analyst revision signals are taking over factor ranking systems

In the U.S. cluster of 58 factors, Up vs Down EPS Revisions rose from rank 10 to rank 1, scoring the highest in the 4-Dimensions Framework. Japan's 12M Fwd Earnings Yield moved from rank 3 to rank 1. Europe's same signal rose from rank 24 to rank 18. This cross-regional ranking shift is not synchronized, implying a divergence in global earnings revision cycles—Japan's structural reforms have driven both the breadth and persistence of upward earnings forecast revisions.

The March 2026 geopolitical shock tested the model's factor timing value

In March, amid rising Middle East tensions and risk of a Strait of Hormuz conflict, the MSCI Japan index fell -12.3% (USD), while the momentum factor suffered massive drawdowns of -7.4% in both long and short legs (described as a 4-sigma crowded position unwind). Over the same period, Japan's low-volatility factor returned +9.0% and the value factor +6.0%, creating extreme divergence among the three. Global energy became the only positive sector that month (Europe +19.45%, U.S. +10.5%, Japan +5.74%). This extreme environment precisely validates the necessity of dynamically feeding macro judgment (geopolitical risk pushing up energy/inflation → rising rates → value outperformance) into factor weight frameworks.

Key Divergences and Risks

Statistical significance risk. Multiple conditional factor return t-statistics have absolute values below 2.0, e.g., the U.S. value factor under rising rates has a t-stat of only 1.7, and Japan's low-vol factor under rising inflation only 1.0. Historical differences could reverse out of sample.

Regional specificity issue. Japan's low-vol factor's long-term negative return is unique globally. If model parameters are not regionally differentiated, Japanese allocations will be systematically wrong. More critically, Japan's momentum factor delivers returns near zero or negative in most macro environments, with very low confidence in any full-sample alpha.

Cascading impact of macro scenario misjudgment. Factor weights are highly sensitive to the direction of interest rates, the economic cycle, and the equity risk premium. If the direction is misjudged (e.g., expecting rates to fall when they actually rise), the model would allocate weights to factors on the opposite side (e.g., favoring growth over value), creating a double negative.

Model decay. Factor leadership changes as cycles turn; analyst inputs require continuous updates, and as crowding increases, factor discrimination declines. Japan's momentum factor's 4-sigma drawdown in March is a clear example.

Valuation or Trading Implications: Systematic Sector Rotation Direction

The current sector rotation framework provides clear long/short directions as auxiliary inputs for QAML timing: U.S. overweight Consumer Staples and Communication Services, underweight Consumer Discretionary and Materials; Japan overweight Utilities and Financials, underweight Materials and Communication Services. This direction is consistent with the macro logic above (rising rate trend, value factor strength, defensive demand) and can serve as a calibration anchor for sector exposure within a beta-neutral strategy.

---

Appendix: Factor Cluster Rankings (Based on 4-Dimensions Framework)

Appendix: Conditional Annualized Returns of Global Factors by Long-Term Rate Direction