Performance comparison
How to choose a backtest benchmark and interpret excess return
A positive historical return does not answer whether a strategy improved on a relevant alternative. A benchmark gives that comparison a reference, but an unsuitable index or mismatched return convention can create an apparent advantage that the strategy did not earn.
Published by FiClaw 产品团队 · Published · Updated
阅读中文版Quick answer
Choose a backtest benchmark before evaluating the result, based on the strategy's opportunity set and research question. Align the test interval, currency, valuation dates and dividend convention, and state the treatment of costs. “Excess return” needs a definition: a difference in cumulative returns, a compounded relative return and regression alpha are different quantities.
Key takeaways
- Choose the reference from the research objective, before seeing which comparison looks best.
- Match dates, currency, dividends and valuation conventions, and disclose cost differences.
- Label the excess-return formula and evaluate risk alongside return.
1. Match the benchmark to the research question
For stock selection within a defined universe, a relevant universe index or a clearly specified passive portfolio may help distinguish selection from general market exposure. A broad A-share index can provide context for an A-share strategy, but it may not represent a small-cap, sector-restricted or tightly constrained opportunity set.
An absolute-return or market-neutral research question may also need a cash or risk-free reference and explicit exposure measures. State how that reference is obtained and calculated. If multiple comparisons are useful, designate a primary benchmark and explain the purpose of each additional one rather than choosing the most favorable result afterward.
| Research question | Potential reference | Question to resolve |
|---|---|---|
| Selection within an index universe | The relevant index or a defined universe portfolio | Do historical membership and weighting conventions match the task? |
| Sector-restricted strategy | A corresponding sector reference, plus broad-market context | Is the result explained by sector exposure rather than selection? |
| Timing or variable market exposure | The underlying passive exposure and a stated cash convention | How are uninvested cash and time out of the market treated? |
| Absolute-return or market-neutral research | A stated cash/risk-free reference and exposure diagnostics | Are financing, leverage and residual market exposure accounted for? |
2. Align the return series before comparing them
Check that strategy and benchmark begin and end on the same effective dates and use compatible valuation frequency, currency and trading calendars. Decide how to align missing observations; silently forward-filling a series or dropping dates can change the comparison.
A price index tracks price changes, while a total-return index also reflects reinvested distributions according to its methodology. Check how dividends and corporate actions enter the strategy series and choose a compatible benchmark series or disclose the mismatch. Also distinguish a published index from an investable tracking product, which can have fees and tracking differences.
- 01Record the exact benchmark identifier, provider, return variant and retrieval date.
- 02Use the same comparison interval and explain any warm-up period excluded from evaluation.
- 03Label the strategy as gross or net of fees and slippage; state benchmark cost assumptions.
- 04Keep annualization and risk-free conventions consistent for the metrics being compared.
3. Say which excess-return calculation the report uses
For the same interval, a strategy return of 10% and a benchmark return of 8% differ by 2 percentage points. The relative wealth return is instead (1.10 / 1.08) − 1, approximately 1.85%. This is an arithmetic illustration, not FiClaw or customer performance. Both calculations can be useful, but they answer different questions and should not share an unexplained label.
Annualized strategy return minus annualized benchmark return is also different from annualizing the relative wealth series. Regression alpha is another measure, dependent on the model, factors, frequency and estimation window. A positive simple excess return should not automatically be described as alpha.
| Label | Definition to disclose | Interpretation |
|---|---|---|
| Cumulative return difference | Strategy cumulative return − benchmark cumulative return | An interval difference expressed in percentage points |
| Relative wealth return | (1 + strategy cumulative return) / (1 + benchmark cumulative return) − 1 | How the strategy's terminal wealth compares with the benchmark |
| Annualized return difference | Annualized strategy return − annualized benchmark return | A difference under stated annualization conventions |
| Regression alpha | Estimated intercept under a stated factor model | A model-dependent estimate, with sampling uncertainty |
4. Document the comparison in the FiClaw research record
Put the benchmark choice and formula in the confirmed strategy spec, then inspect how the actual code and report implement them. FiClaw's code generation and backtest workflow can support this research process; verify the available data series and output conventions instead of assuming a particular index variant or alpha model is built in.
Read excess return with drawdown, volatility, exposure, turnover, costs and performance across subperiods. Outperformance driven by more market exposure or a few exceptional dates needs a different explanation from consistent selection effects. A benchmark comparison supports research evaluation and does not establish future performance.
The comparison template describes evidence to assemble, including separate records when needed. It does not imply that FiClaw automatically supplies every benchmark, factor model, data snapshot or review artifact listed here. The external benchmark and relative-change references explain general methods, not FiClaw API availability or default calculations.
Primary research question: [value] Primary benchmark and reason for choosing it: [value] Identifier, provider and price/total-return variant: [value] Dates, currency, frequency and missing-date handling: [value] Strategy costs and benchmark cost assumptions: [value] Excess-return formula and annualization convention: [value] Additional risk, exposure and subperiod checks: [value] Known mismatches or unavailable benchmark data: [value]
Frequently asked questions
Can every A-share strategy use the CSI 300 as its only benchmark?
The CSI 300 can provide broad context, but it may not match a small-cap, sector-specific or otherwise constrained strategy. Choose a primary reference that fits the opportunity set and research question, and explain any additional broad-market comparison.
Is strategy return minus benchmark return the same as alpha?
No. It is a simple return difference under the stated interval and conventions. Regression alpha depends on an explicit model, factors, observation frequency and estimation window, and should include an assessment of uncertainty.
Why does a price-index versus total-return benchmark matter?
The series treat distributions differently. If the strategy includes dividends but the benchmark excludes them, the comparison can attribute a convention difference to strategy skill. Align the conventions where possible and disclose any remaining mismatch.
Method and product references
Related reading
Turn a research question into a checkable experiment
Start with explicit strategy rules in FiClaw, then review the code, data assumptions and backtest results.