Advanced B · Strategy research
Take eight strategy categories through observation, mechanism, hypothesis, data, signal, backtest, and falsification.
- Chapter 2 · Price discoveryHow can order-flow and capital-flow data measure information's incorporation into price?
- Chapter 4 · TrendHow do you research Time-series Momentum: define signals and explain returns?
- Chapter 6 · Complete tradeHow do you turn a Trade Card into a complete Strategy Object?
- Chapter 9 · Win rate and reward/riskWhat win-rate and reward/risk structures typify trend, mean-reversion, and Carry strategies?
- Chapter 17 · Derivatives and leverageHow do Funding, OI, and Basis combine into Carry and positioning strategies?
- Chapter 18 · Liquidation feedbackHow can liquidation data construct a Signal?
- Chapter 19 · Return sourceWhose money funds each of eight strategy classes, and when does each fail?
- Chapter 20 · HypothesisHow can you state a testable research question and preregister methods before viewing data?
- Chapter 21 · MechanismHow can data test the mechanism itself rather than returns alone?
- Chapter 24 · RefutationHow can you design the test most likely to make a strategy fail, rather than pass?
Begin with the return mechanism
Explain who pays, what risk you bear, and why competition has not erased returns before defining signals. The research chain produces rejectable Strategy Objects. Runnable code, profitable backtests, and professional-sounding names are not acceptance criteria.
- Observation
- Mechanism
- Hypothesis
- Data
- Signal
- Backtest
- Falsification
Eight separate modules
| Module | Research subject | Required counterevidence |
|---|---|---|
| Momentum | Delayed response and continuing capital needs | Fast reversals, ranging, crowded exits |
| Mean Reversion | Repair of temporary deviations | Permanent anchor changes |
| Carry | Compensation for capital, maturity, and risk | Rate reversal, margin/financing runs |
| Relative Value | Relative pricing across two legs | Hedge breaks and leg risk |
| Flow | Incremental future-demand information | Internal transfers, label revisions, publication delays |
| Event | Expectations versus actual events | Priced-in events or revised calendars |
| Volatility | Volatility risk prices and actual paths | Jumps, hedging costs, surface changes |
| Arbitrage | Constraints among executable prices | Asynchronous fills, failed delivery/transfers |
These use Advanced return-mechanism categories. The separate Crypto Strategy Library includes application categories Derivatives and Onchain, which supply data to several modules but do not replace Volatility or Arbitrage.
A shared experimental protocol
Copy Research Memo and Strategy Object before research, freezing the primary hypothesis, market, horizon, costs, and falsification. Modules allow changing inputs, not hindsight relabeling of failures as inapplicable regimes.
Keep at least cash/flat, simple-holding, main-signal-ablation, and cost-stress baselines. Controls need comparable risk and horizons; a leveraged strategy outperforming unleveraged holding does not prove a signal.
Check completeness before training, validation, and final testing. Use then-visible information only and submit after signal arrival. Do not assume free fills at the same candle's closing signal price. Account overlapping positions as portfolio returns, not repeated independent signals. See Systematic research.
Comparing the eight categories
Win rate, payoff, skew, and holding periods arise from definitions and regimes, not fixed category profiles. Trends may take frequent small losses before rare large gains; Carry may look calm before concentrated funding-stress losses. These are shapes to test, not guarantees.
Before curves, list directional exposure, financing dependence, liquidity needs, event gaps, and latency for eight candidates. Two strategies dependent on improving risk appetite are not independent despite their names. Submit this comparison to Portfolio.
Projects and passing criteria
Finish eight exercise cards, then take one through reproducible research with a Research Memo, Strategy Object, and Backtest Report. Keep failed candidates registered. Losing final tests may pass the exercise if you explain the evidence contradicting the mechanism.
Reviewers must reconstruct signals, per-trade/daily equity, costs, remainders, and rejections from inputs/code; locate independent OOS, sensitivities, and a mechanism counterexample; and recalculate net PnL from raw ledgers. Missing any keeps the strategy in Research or Testing, without live promotion.
Liquidation signals need visibility and controls
Liquidation data may describe completed forced trades or estimated maps. Identify sources, venue coverage, first availability, and errors. “Liquidity recovers after extreme selling” and “selling feedback continues” are separate teaching hypotheses, registered before results.
Compare rebound and continuation signals in one event window, adding then-visible OI, Funding, Basis, and depth. Do not group using hindsight highs/lows. Control for broad crashes, ordinary high volume, and data delays to test incremental information. Retain underreporting, estimates, and cost stress. If ordinary trends explain results or true release timing removes effects, reject the liquidation mechanism.