This page summarizes the current best strategy for every tracked market, shows what the sprint is doing now, links the actual QuantConnect in-sample and out-of-sample backtests, and explains why each champion exists. It also separates crypto conservative from crypto aggressive so one mandate can stay deployable while the other keeps hunting high-risk / high-return edge.
The system is designed to be attribution-clean: one change at a time, OOS-first, and always logged.
Each market starts from its current champion. A new version is created with one clean parameter change or one alpha-orthogonal addition. This keeps attribution honest.
The new version is backtested on the in-sample window first. If it cannot clear the market’s minimum IS threshold, it does not earn an OOS run.
If IS passes, the pipeline launches a dedicated OOS test. A strategy only becomes champion if its OOS Sharpe is strictly better than the current champion and it stays inside the market’s degradation and drawdown caps.
Every run is logged into ~/.quant_research/knowledge.db, sprint state is updated, and the champion board is advanced only when the result is truly better.
The cron runs every 30 minutes, works one variant per active market, and respects the QuantConnect node bottleneck. Saturated markets are halted instead of wasting compute.
Crypto now has two separate tracks: a conservative DD-capped track for deployable capital, and an aggressive track that preserves high-risk / high-return moonshots.
Active · near local ceiling
| Repo | ccchoo/quantconnect-trading |
|---|---|
| IS Sharpe | 2.683 |
| OOS Sharpe | 2.038 |
| Max drawdown | 18.1% |
| CAGR | 89.7% |
| Backtests | IS backtest · OOS backtest |
| Iterations run | 88 |
|---|---|
| Champion wins | 5 |
| Current misses | 19 |
| Next version | V2322 |
19 consecutive misses after the current champ. Many variants tie the champion on OOS exactly, which is strong evidence of a local architecture ceiling rather than a broken strategy.
The current equity winner came out of a long line of earnings-quality + momentum + risk-control variants. Earlier work established that naive factor L/S, PCA stat-arb and pairs trading did not survive OOS. The winning family kept the cross-sectional stock-selection core but learned to modulate gross exposure using dispersion and crash controls. V2301 won by lowering the DVS floor from 0.70 to 0.60, allowing deeper de-leveraging in ugly high-dispersion regimes without giving up too much upside in cleaner markets.
Either beat 2.038 OOS cleanly or accept this family is saturated and pivot to a genuinely new architecture. Current focus: small risk-budget / dispersion / crash-scaling nudges. If another miss lands, equity is effectively at ceiling for this design.
Active · still improving at the margin
| Repo | ccchoo/quantconnect_option |
|---|---|
| IS Sharpe | 1.554 |
| OOS Sharpe | 2.144 |
| Max drawdown | 15.7% |
| CAGR | 64.2% |
| Backtests | IS backtest · OOS backtest |
| Iterations run | 82 |
|---|---|
| Champion wins | 16 |
| Current misses | 5 |
| Next version | V477 |
Options is the healthiest research lane operationally: lots of wins, low DD, and still enough room for small OOS improvements.
The options research path started from the original SPY put credit spread champion and then cleaned up a lot of bad assumptions with cloud-forensic reruns. The durable lesson was that many flashy alternatives failed, while disciplined structure tuning kept compounding. The recent winning streak came from stacking small, attributable changes: higher take-profit, then a wider covered-call OTM band. V470 is the current best expression of that family.
Continue micro-optimizing the profitable family: covered-call OTM distance, DTE windows, volatility targeting and execution rules. This market still shows incremental improvement instead of full saturation.
Active · DD-first champion installed
| Repo | ccchoo/quantconnect-crypto-trading |
|---|---|
| IS Sharpe | 0.923 |
| OOS Sharpe | 1.075 |
| Max drawdown | 15.2% |
| CAGR | 40.5% |
| Backtests | IS backtest · OOS backtest |
| Iterations run | 55 |
|---|---|
| Champion wins | 3 |
| Current misses | 7 |
| Next version | V10668 |
This is the “institutional-grade” crypto lane. It intentionally gives up triple-digit moonshot behavior in exchange for survivable drawdowns.
This champion exists because the research process deliberately changed objective. The old crypto leaders produced huge Sharpe and CAGR, but at 60%+ drawdowns. Once the DD ceiling was tightened to 40%, the whole old family failed qualification. The restart discovered that equity-style 12-15% vol targeting was far too conservative for crypto. The solution was not simply “less leverage”; it was regime-aware leverage. V10661 emerged from that redesign: vol-percentile regime switching + EMA trend + 20d momentum + a hard 15% portfolio drawdown stop.
Tune vol-regime thresholds, test whether a carefully filtered third asset can lift Sharpe, and keep the hard DD discipline intact. The priority is not moonshot return — it is keeping OOS DD under 40% while preserving deployable edge.
Active · high-risk / high-return track
| Repo | ccchoo/quantconnect-crypto-trading |
|---|---|
| IS Sharpe | 4.671 |
| OOS Sharpe | 6.673 |
| Max drawdown | 67.2% |
| CAGR | 809.9% |
| Backtests | IS backtest · OOS backtest |
| Iterations run | 5 |
|---|---|
| Champion wins | 3 |
| Current misses | 0 |
| Next version | V10804 |
This lane is intentionally separated from conservative crypto so the dashboard can preserve both the moonshot and the deployable versions of the same market.
When the conservative crypto track tightened DD to 40%, the old monster winner should not have been forgotten — it should have been reclassified. That created the aggressive track. The baseline was V10602, then the research simply followed the strongest live pattern: BOOST 1.35→1.45 won, then 1.45→1.55 won, then 1.55→1.65 won. V10802 is the current endpoint of that chain. This is not accidental overfitting-by-chaos; it is a deliberately separate mandate optimized for high risk / high return.
Keep walking the winning direction carefully while staying under the relaxed 70% DD cap. The next wave is likely more BOOST / regime tuning and potentially asset-expansion variants that preserve the acceleration edge.
Halted · saturated at daily resolution
| Repo | ccchoo/quantconnect-forex |
|---|---|
| IS Sharpe | 0.394 |
| OOS Sharpe | — |
| Max drawdown | 25.7% |
| CAGR | 11.6% |
| Backtests | IS backtest · OOS link unavailable (no qualifying OOS champion) |
| Iterations run | 22 |
|---|---|
| Champion wins | 0 |
| Current misses | 20 |
| Next version | FX_V60 |
Forex is paused for good reason. The current next plan is a research pivot, not another parameter nudge.
Forex is the cleanest example of the research process refusing to hallucinate edge. The old daily OANDA majors family peaked around IS Sharpe 0.35-0.40. New variants kept failing either IS or OOS. The sprint was halted after 20 consecutive misses. FX_V11 remains the best documented daily-resolution baseline, but the real lesson is architectural: the current daily family is saturated and should not absorb more compute unless the design changes materially.
Do not keep grinding daily Donchian/SMA/ATR variants. The next serious attempt should pivot to intraday/session mean reversion, carry with real macro inputs, cross-sectional FX baskets, or a proper regime-switching framework.
What the research program has learned so far.