What things to consider while back testing?

<p id="isPasted">There are various factors on which the back testing strategies may depend. Few of them are sighted below :</p><ul><li>Past data of market behaviour.</li><li>Model refining, so that it best fits historical data, without accounting for the fact that future conditions may be different.</li><li>Past datasets on an adverse market event or sentiment</li><li>Sufficient datasets to likely produce models that would account for a wide variety of market conditions</li><li>A trading strategy that should work well on several datasets from one market (eg forex) as well in another market (eg shares)</li><li>Strategies that should be tested well in both bullish market and bearish environment, and vice versa.</li></ul>

<p id="isPasted">Backtesting is a critical bridge between a trading theory and live execution, but its results are only as good as the realism you build into the simulation.&nbsp;</p><p><strong>Data Integrity &amp; Quality</strong></p><p>The foundation of any backtest is high-quality, "clean" historical data.&nbsp;</p><ul><li>Completeness: Ensure there are no gaps in historical price or volume data, especially around high-volatility events like earnings or macro announcements.</li><li>Survivorship Bias: Use datasets that include companies that were eventually delisted, sold, or went bankrupt. Testing only on currently existing stocks artificially inflates results by ignoring the "losers".</li><li>Corporate Adjustments: Data should be adjusted for splits, dividends, and rights issues to maintain historical continuity.&nbsp;</li></ul><p><strong>Real-World Friction</strong></p><p>Most backtests fail because they assume perfect, frictionless execution.&nbsp;</p><ul><li>Transaction Costs: Always include brokerage commissions, exchange fees, and taxes. Even small fees can turn a profitable high-frequency strategy into a losing one over time.</li><li>Slippage: Factor in the difference between your expected price and the actual execution price. A common fix is to deduct 0.1% to 0.5% per trade to simulate these real-world losses.</li><li>Liquidity Constraints: Ensure your simulated trade sizes are realistic compared to the average trading volume of the asset. Large orders in thin markets can move the price against you.&nbsp;</li></ul><p><strong>Strategy Robustness &amp; Validation</strong></p><p>To ensure your strategy isn't just "lucky" with past data, use rigorous validation techniques.&nbsp;</p><ul><li>Avoid Overfitting (Curve-Fitting): Do not tweak parameters so much that they only fit the noise of a specific historical period. Simple strategies with fewer parameters (typically 3–5) are often more robust.</li><li>Out-of-Sample Testing: Split your data into two sets. Use the first set (in-sample) to build the strategy and the second, untouched set (out-of-sample) to validate if it still works on "unseen" data.</li><li>Diverse Market Conditions: Test across different "regimes"—bull markets, bear markets, and sideways/range-bound periods. A strategy that only works in one environment will likely fail when the market shifts.&nbsp;</li></ul><p><strong>Key Performance Metrics</strong></p><p>Look beyond just the "Total Profit" to understand the strategy's risk profile.&nbsp;</p><ul><li>Maximum Drawdown: The largest peak-to-trough decline in your capital. This tells you if you have the psychological stomach to survive a losing streak.</li><li>Risk-Adjusted Returns: Use the Sharpe Ratio (profit per unit of risk) or Sortino Ratio (focusing on downside risk) to compare the strategy against a benchmark.</li><li>Sample Size: Ensure you have enough trades for statistical significance. Aim for at least 100+ trades over several years to ensure results aren't random fluctuations.&nbsp;</li></ul>