Define Mcginley Dynamic and the exact backtest target
Mcginley Dynamic is a moving-average style method that produces a dynamic line intended to adapt to changes in trend speed. Before you backtest it, define two things precisely:
- The input series (for example, which price field you use: midpoint, bid, ask, or close).
- The backtest objective. A backtest is not just “compute the indicator.” It must specify what decision or measurement you evaluate, such as prediction quality of a chosen relation or performance of a rule.
If your goal is indicator quality rather than trading, choose a measurable target such as how often the line is above/below a reference series, or whether a calculated slope matches later movement. Whichever target you choose, state it up front so the evaluation does not drift after you see results.
Mechanics: choose data, resampling, and rules that you can reproduce
A responsible backtest starts with reproducible mechanics.
Data granularity and alignment
- Use one consistent timeframe (e.g., hourly bars) and document whether the indicator is computed from bar closes or from intrabar timing approximations.
- If you use end-of-bar prices, ensure the signal (or measurement) is timestamped so you do not accidentally use information from the future.
Parameter assumptions Mcginley Dynamic typically involves a parameter that controls responsiveness (commonly denoted in descriptions as a constant). Even if you keep the indicator formula fixed, you must state:
- The chosen parameter values.
- Whether the same values are used across all test periods.
Costs model Backtests that ignore frictions can produce misleading results. Include at least:
- A proxy for spread/slippage as a cost per entry/exit.
- A note on whether costs are applied at the bar you act on.
Position and execution rules If you test trading rules, define:
- When orders are considered filled (at close, next open, or at an approximation you document).
- How you handle multiple signals inside one timeframe.
Evidence and example approach: compare in-sample vs. out-of-sample
Because historical relationships do not guarantee future results, a careful method separates model selection from evaluation.
A practical structure
- In-sample (training) period: pick parameter values and the evaluation rule.
- Out-of-sample (testing) period: apply those fixed choices without further tuning.
- Repeat across several time windows using a walk-forward schedule.
This reduces the chance that good results come from overfitting. If you re-tune parameters after seeing test performance, you weaken the meaning of the test.
Bias controls
- No look-ahead: ensure every value used to make a decision would have been available at that time.
- Fixed rule set: do not change the objective after observing outcomes.
- Consistent data treatment: handle missing data the same way in all periods.
Stress tests Run additional checks where outcomes are likely to be fragile:
- Regime shifts (slow trend vs. fast trend environments).
- Sudden spread widening and cost assumptions.
- Different liquidity periods, if your dataset distinguishes them.
Limitations and failure modes to expect
Even with careful controls, backtesting can fail in predictable ways.
Transaction cost sensitivity Many indicator-based strategies have edge sizes that can be smaller than realistic costs. Small changes in spread/slippage assumptions can flip conclusions.
Data quality and construction issues Backtests can be sensitive to how the price series is built (midpoint vs. tradeable quotes). Missing bars or timestamp misalignment can create artificial patterns.
Overfitting to history If you try many parameter values and selection criteria, you may “find” a pattern that fits the past but does not generalize.
Non-stationarity Market dynamics change over time. An indicator that responds well in one regime can behave differently in another, especially if the market’s volatility or trend structure shifts.
Verification and what to do next
To verify your backtest is responsible, check that you can independently explain and reproduce it:
- You can state data choices, indicator mechanics, parameter values, and execution timing.
- You can show that results come from out-of-sample evaluation, not only from in-sample tuning.
- You can list at least one scenario where the method is likely to break (for example: higher costs, different price inputs, or regime changes).