Improve Trading StrategyNovember 4, 202616 min read

How to Improve Your Trading Strategy With Evidence

Improve your trading strategy step by step: write testable rules, backtest with realistic costs, avoid overfitting, review trades and validate before risking capital.

To improve a trading strategy, turn it into written rules, test those rules on historical data with realistic costs, change one thing at a time, and confirm the change holds on data the strategy has never seen. Then compare live results with what the test promised. Skipping any step usually means you are tuning noise, not improving anything.

This guide walks through that loop in order. It covers what to measure, which changes are worth testing, how to avoid fooling yourself, and how to size risk once the numbers look good.

TL;DR

Write every entry, exit, risk and filter rule so a stranger could apply it without judgment. You cannot improve what you cannot test.
Measure win rate, profit factor, maximum drawdown and average trade together. No single number tells you whether a strategy is good.
Include fees, slippage and liquidity in every test, and change one rule at a time.
Check each change on data the strategy was not tuned on. A better backtest alone proves little.
A backtest is hypothetical. Compare live results with the test's expectations and size positions so one bad run cannot end the account.

Start with rules you can actually test
Get a baseline before you change anything
Read the numbers that matter
Review the trades, not just the totals
Change one rule at a time
Guard against overfitting
Validate before real money
Check strategies you copy from videos
Fit position sizing to the strategy
A worked improvement cycle
Improve faster with Backtestify
FAQ
Sources

Start with rules you can actually test

Most strategies that "need improvement" are not strategies yet. They are a feeling about a chart pattern plus some discretion on the day. Discretion makes results impossible to compare, because you never know whether a loss came from the idea or from the way you applied it.

A professional research shop describes the standard: every trading decision is specified in advance, so the result is reproducible and the reasoning can be reviewed. That is the bar to aim for, even if you trade by hand.

Write down four blocks:

Entry: the exact condition that triggers a trade. "Price closes above the 50-day moving average while the 14-day RSI is above 50" is testable. "Strong momentum" is not.
Exit: where you take profit and where you admit you are wrong. Include time-based exits if you have them.
Risk: how much you risk per trade and how large a position that allows.
Filters: the conditions under which you do not trade at all, such as low-volume sessions or the minutes around major data releases.

Also write one sentence on why the pattern should exist. If you cannot say who is on the other side of your trade and why they keep losing to you, you probably found a coincidence. Strategies with a reason behind them tend to survive changes in market conditions better than ones that merely fit a chart.

If you want the mechanics of turning those rules into a first test, our step-by-step backtesting guide covers data, entry and exit definitions, and reading the report.

Get a baseline before you change anything

Improvement means "better than what," so you need a baseline. Run the strategy exactly as written over a long enough history to include different market regimes: trending, choppy, calm and panicked. Save the results. Every change you test later gets compared with this run, not with your memory of it.

A baseline needs enough trades to mean something. A strategy with 20 trades can show a 70% win rate by luck alone. As a working habit, treat anything under a few dozen trades as an anecdote, and look for sample sizes in the hundreds before you trust a pattern. Intraday strategies produce trades quickly, so our guide to intraday backtesting is a good companion if you trade short timeframes.

Put realistic costs in the baseline

The most common reason a backtest looks great and a live account does not is cost. Research on investment model validation from the CFA Institute's research foundation stresses that market impact, transaction costs, fees and liquidity constraints must be accounted for, or a strategy may prove unprofitable in live trading.

In practice that means:

Commissions and exchange or regulatory fees for your actual broker and instrument.
Slippage: the gap between the price you expected and the price you got. It is larger for market orders, thin stocks and fast markets.
Spread on every round trip, which matters most for scalping and for small-cap names.
For options, fills at the bid and ask rather than at the midpoint. See how to backtest an options strategy with realistic fills.

High-frequency styles are the most sensitive. A scalping edge of a few cents per share can vanish after costs, which is why a dedicated approach to scalping strategy backtests is worth reading before you trust one.

Read the numbers that matter

Once the baseline runs, look at several metrics together. Each answers a different question.

Metric	What it tells you	What to watch for
Win rate	Share of trades that made money	Meaningless alone. A 35% win rate can be profitable if winners are large.
Profit factor	Gross profit divided by gross loss	Above 1 means profitable before you consider risk. Very high values on few trades are suspicious.
Average win vs. average loss	Payoff ratio	Pair it with win rate to see the expectancy per trade.
Maximum drawdown	Largest peak-to-trough drop in equity	Ask whether you would really keep trading through it.
Number of trades	Sample size	Too few means the other numbers are unreliable.
Time in drawdown	How long recoveries take	Long flat stretches break discipline even when the math works.

Expectancy ties the first three together. If you win 40% of the time, average a $300 win and a $100 loss, your expected result is 0.4 × $300 − 0.6 × $100 = $60 per trade before costs. Change the win rate to 30% with the same payoffs and expectancy falls to $20. That is how a small shift in one number can change a strategy from comfortable to marginal.

The point of the table is to find the weak spot. A strategy with a decent profit factor and a 45% drawdown has a risk problem, not an entry problem. One with a modest drawdown and a profit factor barely above 1 has an edge problem, where costs will likely finish it. Diagnose first, then pick what to change.

Review the trades, not just the totals

Totals hide patterns. Go through the list of individual trades and sort them a few ways:

By outcome: what do the biggest losers have in common? Often it is one condition, such as entering right before earnings or trading against a higher-timeframe trend.
By time: do results cluster in certain hours, weekdays or months?
By market regime: does the strategy make all its money in rising markets and give it back in sideways ones?
By holding period: do trades that run longer than your average behave differently?

Each pattern is a hypothesis, not a fix. "Losers cluster in the first 15 minutes" suggests a time filter, and you then test that filter like any other rule. Professional shops apply the same discipline after launch: they compare realized performance such as fills, turnover and hit rate against the expectations set during research.

Keep a plain journal of what you changed and why, with the date and the result. It stops you from quietly re-testing the same idea a fifth time until it "works."

Change one rule at a time

When results disappoint, the temptation is to rewrite everything. Resist it. If you change the entry, the stop and the filter together, you cannot tell which change helped and which hurt.

A sensible order of changes:

Remove or add a filter. Filters that block obviously bad conditions are usually the cheapest improvement.
Adjust the exit. Compare a fixed stop with a volatility-based stop, or a fixed target with a trailing exit. Exits often matter more than entries.
Refine the entry. Only after the first two, and only with a reason beyond "it fits the chart better."
Tune parameters last. Moving average lengths and thresholds are where overfitting lives.

For each change, rerun the full test with identical data and costs, then compare against the baseline on the whole table above, not only on profit.

A side-by-side view helps. Backtestify's reports show an original strategy next to an improved version on recent data, so you can see whether the change helped across the board or traded one weakness for another. You can browse published examples in the strategy library, including results for strategies that did not hold up.

Guard against overfitting

Overfitting is the central danger of improvement work. Academic material on portfolio optimization describes it as developing a strategy that performs well on a backtest by exploiting random historical patterns unlikely to recur. Every extra rule and every tuned parameter gives the strategy another chance to memorize the past.

Warning signs:

Performance collapses when you nudge a parameter slightly, say a 20-day lookback works but 18 and 22 do not.
The improved version needs many more rules than the original.
The gain came almost entirely from a handful of trades.
You tried dozens of variants and kept the best one without counting how many you tried.

Practical defenses:

Hold out data. Develop on one period, then test the final version once on a later period you did not touch. Treat that one test as precious.
Prefer plateaus to peaks. Choose parameters from a broad region where results are similar, not a single spike.
Keep rules few and reasoned. Each rule should have a reason that is not "the backtest liked it."
Test across instruments and regimes. A rule that works only on one ticker in one year is a story, not an edge.

We go deeper in our guides to optimizing a strategy without overfitting and to spotting and stopping overfitting in backtests.

Validate before real money

The purpose of validation is not to produce the prettiest equity curve. It is to find the ways the strategy can fail while failure is still free. Backtesting alone cannot do this, because every backtest is hypothetical. An industry guidance document on backtesting says such results are hypothetical, not actual performance, and cannot guarantee future results, and that they should be kept distinct from live trading history.

That is also the honest answer to anyone hoping to predict the next price move. A backtest tells you how a rule set would have behaved, with what frequency of wins and what depth of losses. It supports probabilities and ranges, not certainty.

A sensible validation sequence:

Out-of-sample test on data you reserved.
Forward test on live data with no real money, or the smallest size your broker allows, and with every signal logged before the outcome is known. Our guide to forward testing with auditable results lists five checks to keep that record honest.
Compare to expectations. Define in advance what range of win rate, average trade and drawdown counts as normal for this strategy. If live results fall well outside it, pause and investigate before adding size.
Scale gradually. Increase size only after a meaningful number of trades behaved as the test suggested.

A broader view of this mindset is in our piece on evidence-based trading, and a regulator-style checklist is in backtesting like a regulator.

Check strategies you copy from videos

Many traders start from a setup they saw on YouTube. That is a fine source of ideas, and a poor source of evidence. A video shows a few hand-picked examples. It rarely shows every signal over several years, the losers, or the costs.

Treat a creator's rules like any hypothesis: translate them into precise entry and exit conditions, run them over a long history, and look at the full distribution of results. You may find the strategy holds up, holds up only in certain conditions, or does not hold up at all. All three outcomes are useful and cheaper than discovering them live. Backtestify's library includes examples such as the Tori Trades trendline strategy, tested over real market data.

Fit position sizing to the strategy

A strategy can have a real edge and still wipe out an account if each trade risks too much. Improving risk often does more for your results than improving entries.

One popular convention is the 3-5-7 rule. A published explanation describes it as a position-sizing convention that limits risk per trade to 3%, per underlying to 5% and in aggregate to 7% of account equity. Other write-ups define it differently, with 3% risk per trade, 5% across open positions and winners aiming to be at least 7% larger than losers. Because versions differ, treat it as a starting framework to adapt, not a law, and test the exact version you choose inside your backtest.

Two versions of the 3-5-7 sizing rule

Many active traders risk well under 3% per trade. Whatever number you pick, run the backtest with that sizing included, then look at the maximum drawdown it produces. If that drawdown is larger than you could live with, lower the risk per trade and test again. For low-priced, thinly traded stocks where slippage and gaps are larger, a risk-first guide to position sizing for penny stocks shows how those limits translate into share counts.

Execution matters too. The CFA Institute notes that a strategy's goal is to balance expected costs, risks and alpha, and that execution style should match how urgent the trade is. For retail traders that mostly means choosing limit versus market orders deliberately and modeling that choice in the test.

A worked improvement cycle

Here is the loop on a hypothetical example, with made-up numbers for illustration.

A trader buys breakouts above a 20-day high on a liquid ETF. Baseline over ten years with costs: 180 trades, 38% win rate, profit factor 1.15, maximum drawdown 22%. Profitable, but thin.

Review: many losers occur when the broader market is below its 200-day average.
Change one rule: skip breakouts when the index is below that average. Rerun. Trades fall to 120, profit factor rises to 1.4, drawdown falls to 14%.
Check the plateau: test 150-day and 250-day averages. Results are similar, so the rule is not hanging on one value.
Hold-out test: run the final rules on the two most recent years, untouched until now. Profit factor is 1.25, lower than in development, which is normal and still positive.
Forward test with small size and compare with the expected range.

Notice what did not happen: no new indicators, no 15 parameters, no cherry-picked period. The improvement is one reasoned filter, verified out of sample.

Improve faster with Backtestify

Doing this by hand in a spreadsheet works, but it is slow, and it is easy to skip the discipline. With Backtestify you describe the strategy in plain English, run it on real historical market data, and get a trade-by-trade report with win rate, profit factor and maximum drawdown. You can then improve the rules and compare the original and improved versions on recent data, and export the result in TradingView Pine Script format.

The Free plan lets you get started, with details on the billing page. Pro, at $29 per month or $190 per year, adds unlimited backtesting, improvement and forecasting. You can see both on the Backtestify home page.

Backtestify

FAQ

How do I know if my strategy is actually improving?

Compare each version with your saved baseline on the same data and the same costs, looking at profit factor, drawdown and trade count together. Then confirm the better version on data you held back. If the gain disappears out of sample, it was probably overfitting.

How many trades do I need before trusting a backtest?

There is no universal number, but a few dozen trades is too few to say much and several hundred is a far better foundation. More important than the count is variety: the sample should cover rising, falling and sideways markets.

Can a backtest predict the next price movement?

No. Backtest results are hypothetical and cannot guarantee future results. What a backtest can show is how often a rule set won, how big its losses were, and how bad a bad stretch has been, which lets you plan with probabilities.

What should I change first when a strategy loses money?

Start with costs and risk, then filters and exits, and leave entries and parameter tuning for last. Change one thing per test so you can tell what helped.

Is the 3-5-7 rule a good risk rule?

It is a reasonable framework for capping risk per trade and in total, but sources define it differently. Pick one version, include it in your backtest and check that the resulting drawdown is one you could tolerate.