Forecast first, rank second, allocate last
F2R is a live cross-asset Portfolio Strategy System that separates forecasting from portfolio choice. Conventional supervised models and Chronos-2 produce distinct forecasts, which are brought into a common cross-sectional ranking process before the portfolio is formed.
F2R does not treat raw forecast magnitudes as portfolio weights. Forecasts are first converted into comparable relative-opportunity ranks; a separate allocation rule then turns the final ordering into the monthly multi-asset portfolio. This keeps the forecasting logic separate from the rules that determine the portfolio.
Heterogeneous forecasts, one common ranking process
Combine distinct forecasting views
Conventional supervised models and Chronos-2 process price history through different representations before their forecasts enter the common ranking process.
Normalize forecasts through rank consensus
Heterogeneous outputs are translated into a comparable cross-asset ordering rather than mixed as raw forecast scales or used directly as portfolio weights.
Translate ranks into the portfolio target
A separate allocation rule maps the final ordering into the monthly model portfolio under a defined timing, turnover, cost, and accounting process.
Live Completed Performance
Updated with each F2R release. The chart and statistics show completed performance for the current F2R model. Current MTD and Preview are kept separate and are excluded until the holding month is complete.
Cumulative Performance
2017-05-01 → 2026-09-30 · completed through 2026-09-30 · current MTD excluded
Recent Completed Monthly Returns
Recent 12 completed months · 2025-10 → 2026-09 · current MTD and Preview excluded
| Series | Cumulative | CAGR | Vol | Sharpe | MDD | Calmar |
|---|---|---|---|---|---|---|
| F2R · current canonical model | 366.91% | 17.78% | 17.68% | 1.02 | -21.97% | 0.81 |
The model-adoption study remains a separate retrospective record
Chronos-2 adoption evidence remains citable as a frozen research counterfactual. It is preserved for research provenance and no longer serves as the primary performance evidence for the live F2R System page.
Official and Preview are shown as different operating states
The standalone F2R dashboard publishes the current official monthly target alongside a separately labeled intramonth preview. The preview is indicative only: it may change before month-end, is not executed, and is kept outside completed performance. Historical research results and current operation are also shown separately.
Research papers document distinct stages of the F2R research record
The earlier information-set study tests whether historical-vintage macro information adds decision value beyond price information under matched forecasting, ranking, portfolio, and timing rules. The newer model-adoption study keeps the portfolio-side price-data domain and portfolio interface fixed while asking how Chronos-2 changes rankings, Top-4 decisions, and historical portfolio outcomes.
Both papers document earlier stages of the F2R research program. They show the research path that informed the F2R design family, while the live system is maintained separately. Model-combination settings reported in a paper describe that research configuration and should not be read as the current production recipe.
Methods used in this system
These companion articles explain methods used directly in the system or needed to interpret its validation and outputs.
Out-of-Sample Forecast Evaluation
Defines the out-of-sample information boundary shared by F2R's heterogeneous forecasting layer before conventional ML and Chronos-2 views are converted into relative ranks.
Rolling vs Expanding Windows
Explains the time-ordered estimation design used by F2R's conventional supervised models; Chronos-2 is evaluated through a separate inference path rather than sharing the same fitting window.
Look-Ahead Bias and Data Leakage
Defines leakage controls for F2R's point-in-time inputs, supervised-model training, pretrained-model evaluation, current-model reconstruction, and rank-based portfolio validation.
Portfolio Backtesting and Rebalancing
Explains how F2R's rank-based monthly target becomes a continuous holding-period wealth path under explicit execution timing and between-rebalance weight drift.
Turnover, Transaction Costs, and Net Performance
Explains the drift-aware turnover and transaction-cost conventions used to translate F2R's monthly model-portfolio changes into net performance.
Benchmark-Relative Portfolio Evaluation
Defines benchmark-relative measures used to judge the portfolio value of F2R's heterogeneous Forecast-to-Rank decision process beyond same-universe and external comparators.
Regularized Linear Models: Ridge, Lasso, and Elastic Net
Defines the regularized linear-model family that anchors F2R's conventional supervised forecasting lane, including scaling and tuning controls for correlated predictors.
Time-Series Model Selection and Hyperparameter Tuning
Defines chronological model-selection and hyperparameter-tuning discipline used when F2R compares conventional supervised candidates without leaking future information.
Cross-Sectional Ranking, Top-K Selection, and Hold Buffers
Explains the rank-to-portfolio translation used after heterogeneous forecasts enter a common rank-based process, including Top-K selection, persistence, and the distinction between ranks and portfolio weights.
Forecast Combination and Model Averaging
Explains how heterogeneous predictive views can be combined or aggregated without confusing forecast weights, forecast scales, rank aggregation, and downstream portfolio weights.
Pretrained Time-Series Foundation Models
Defines the evaluation boundary for Chronos-2 and other pretrained time-series foundation models, separating inference-time information from pretraining knowledge and historical-evaluation claims.