Backtesting
Lookahead bias
A research error where an analysis uses information that would not have been available at the simulated decision time.
- Timestamp alignment, revised data, and future labels are common sources.
Related: point-in-time data, data leakage, out-of-sample
Backtesting
Out-of-sample
Data reserved for evaluation after rules or parameters have been chosen on separate development data.
- Repeated tuning to the same evaluation set makes it part of the research process and weakens the evidence.
Related: walk-forward analysis, overfitting, validation
Risk
Tail risk
Exposure to rare but severe outcomes that can dominate long-term results or survival constraints.
- Tail risk is often underrepresented by average return or ordinary volatility alone.
Related: stress test, drawdown, ruin
Portfolio
Concentration
The degree to which exposure is dominated by a small number of holdings, factors, sectors, or assumptions.
- Concentration can be intentional, but it should be measured and justified.
Related: diversification, exposure, allocation
Methods
Hypothesis
A testable claim that states what relationship or behavior the research expects to observe.
- A useful hypothesis includes a measurable outcome and a condition that would reject or weaken the claim.
Related: pre-analysis plan, primary metric, falsifiability
Methods
Pre-analysis plan
A written note that defines the question, sample, metric, and decision rule before reviewing the final result.
- The plan reduces hindsight editing and makes exploratory work easier to label honestly.
Related: hypothesis, research journal, p-hacking
Portfolio
Rebalancing
The process of returning allocations toward defined targets or adjusting them under documented rules.
- A rebalancing rule should specify cadence, tolerance bands, and exception handling.
Related: allocation, drift, turnover
Backtesting
Survivorship bias
A research error caused by testing only observations that survived to the current sample.
- Ignoring failed, delisted, or removed observations can make historical results look stronger than they were.
Related: universe selection, sample period, benchmark
Educational boundary
This content is for educational and technical research purposes only. It is not financial advice, investment advice, trading advice, tax advice, or legal advice. Backtests and examples may contain errors or omissions. Past performance does not guarantee future results. Always test code in a safe environment before using it with real accounts or live trading systems.
InQuantWeTrust publishes educational content for quantitative research and analytical methods. It does not provide personalized advice, trade recommendations, managed services, or guaranteed outcomes.