20 years of what data? daily or tick? 20 years of daily is only 2500 points of data that is nothing. details matter in trading "business"
What the harm in testing over 20 years data. Today computers are ultra fast. Some people just want to live in Ignorance I guess.
20 years of what data 1 minute or 1 day or 5 minute or what? days have nothing to do with it - that is child talk - your not thinking Why “20 Years of Data” Can Be Misleading 1. Daily Data (20 Years) One trading year ≈ 252 trading days. 20 years × 252 ≈ 5,000 data points. For a system with multiple parameters, 5,000 samples is often too few to be statistically robust. Example: If your system trades only 50 times per year, that’s just 1,000 trades in 20 years → not enough to prove robustness. 2. Intraday Data (20 Years, 5-Minute Bars) One trading day ≈ 6.5 hours (U.S. equities) = 390 minutes. 390 ÷ 5 = 78 bars per day. 252 trading days × 78 bars ≈ 20,000 bars per year. 20 years × 20,000 = ~400,000 data points. That’s 80x more information than daily bars. 3. Why It Matters The number of trades and data points is what determines statistical power, not just “years.” A system tested on 20 years of daily bars could be overfit and still “look good,” because it hasn’t truly been stressed with enough observations. A system tested on 20 years of intraday (5-minute) bars has hundreds of thousands of samples, making it far more reliable. ✅ How to Explain It Simply When someone says “I tested 20 years of data”, you should ask: “20 years of what frequency?” “How many trades did the system actually take?” “How many independent data points were used to validate it?” It’s not about how many years you tested — it’s about how many meaningful data points you tested on.
Test with all the valid data that you have, and the more valid data you have the better. Why wouldn't anyone want to do that? Unless they have a system that is not robust and is designed to work only on recent data.
you will be educated if you stay here on this forum in spite of yourself 1. Think in Terms of Trades, Not Just Bars What really matters is how many independent trades your system takes. If your strategy only triggers 10 trades per year, then: 20 years = 200 trades total → ❌ not enough to trust. If your system triggers 500 trades per year, then: 20 years = 10,000 trades total → ✅ much more reliable. A general rule of thumb: Minimum ~1,000 trades to start drawing conclusions. 5,000+ trades preferred for robust systems. 2. Statistical Power & Overfitting Risk Each degree of freedom (parameter) in your model requires more data to validate. If you test 5 parameters on a system with only 500 trades, you’re very likely just fitting noise. More trades = less risk of curve fitting. 3. Data Granularity Matters Daily bars over 20 years = ~5,000 data points. Often too few for short-term trading systems. Hourly bars = ~40,000 points over 20 years. 5-min bars = ~400,000 points over 20 years. Much stronger test base. 4. General Quant Guidelines <500 trades → exploratory only, not valid for real money. 1,000–2,000 trades → early validation stage, but still fragile. 5,000+ trades → decent confidence. 10,000+ trades → strong statistical reliability. ✅ Bottom line: A valid test usually means at least 1,000 trades, and ideally 5,000–10,000+ trades across a wide range of market conditions. It’s less about the number of years and more about the number of trades and unique market regimes covered.
Hey Numza! 5 day back test FOR A GREAT SYSTEM? Are you trolling for no nothings, or are you part of the group.
You have the magical, Holy Grail, ability to double your money every 20 days and you decided to give it away for a small fee, Life is a poker table... if you don't know who the sucker is in an exchange environment... it's you.