Altcoin alpha research: test the process, not just the backtest
A research workflow for controlling data leakage, selection bias, unrealistic execution, and the temptation to confuse a fitted chart with a durable edge.
07 / EVIDENCE BEFORE CONVICTION
An attractive chart is a starting point for questions—not proof of a durable edge.
Explain the economic mechanism before choosing the parameters.
Distinguish quoted spreads and gross returns from net economics.
Control data leakage, selection bias, and unrealistic execution.
An alpha-strategy claim should identify what the strategy does, why someone might be paid to do it, and which events create losses. Broad directional exposure during a rising market is not by itself evidence of skill. A model should be evaluated against a relevant baseline using consistent dates, currency, and costs.
Separate an economic hypothesis from historical evidence and from live operational results. Each answers a different question. The research library does not provide trading signals, model portfolios, or performance promises.
A simplified dated-futures carry structure buys spot and sells a matching future. If the future is priced above spot, the entry spread is a starting point for analysis. Financing, execution, settlement, and the resources needed to maintain both legs determine the economic result.
The BIS paper Crypto carry examines Bitcoin and Ether spot-futures differences and limits to arbitrage. Its findings are not a current return quote or evidence of an available opportunity in every altcoin.
A perpetual funding strategy needs a different analysis from a dated basis trade. Funding can vary over the holding period, while a dated trade has a specified settlement horizon. Calling both “market neutral” obscures this distinction unless the mechanism is explained.
Use only information that would have been available at each simulated decision. Record listings, delistings, publication times, and transformations. A universe consisting only of today's survivors can omit exposures a historical investor might have held.
Record trial settings and rejected variants. A period used to tune a model is not untouched evidence for the final design. Chronological validation helps keep later information out of earlier decisions.
Include contract units, realistic delays, transaction costs, and financing. A negative weight in a spreadsheet is not proof that a short position could have been opened at that time and price.
Review attribution, turnover, drawdowns, and collateral demands. Investigate whether a few trades or one favorable market period explain the apparent success. Sensitivity tests should challenge material assumptions rather than add decorative precision.
A useful research outcome can be rejecting the strategy, identifying a data gap, or choosing a simpler model. The long-form articles below emphasize this evidence trail. None of the hypothetical examples establishes future profitability, and a convincing research presentation does not remove investment risk.
Build your understanding, one useful question at a time.