Trading statistics
Which numbers matter: win rate, risk reward, and Sharpe ratio
The trading statistics that decide a backtest are expectancy, profit factor, the risk-reward ratio you actually collected, maximum drawdown, and a Sharpe ratio read on a long sample. Win rate is the number people quote, and it is the one that hides the session. A 62% win rate still loses money when winners are scratched at break-even and losers run to a wide stop. A 40% win rate with a 2R average winner and a 1R average loser has expectancy of 0.20R per trade. Hour, day, trade count, market condition, forecast, stop distance, and early break-even are how you find out whether that expectancy is real. None of those splits exist until a journal has kept every row for a long time.
TestingView writes that journal while you replay the TradingView layout you already trade. Your indicators stay on the chart. Future candles stay hidden. The statistics are measured from the trades you actually took.
Key takeaways
- Expectancy, profit factor, realized risk-reward (RR), drawdown, and Sharpe ratio decide if the account paid. Win rate is the trailer.
- Results by hour and by day show the part of the session that funds you, and the part that donates it back.
- Trades per day and trades per week are a pace. Pace changes costs, fatigue, and how fast a sample becomes real.
- The same setup in a trend and in a chop is two trades. Tag the market condition or the average will lie.
- A forecast extends the record you already have. It does not predict next month's price.
- Optimal stop and target studies ask what your existing trades would have done at other distances.
- Break-even too early shows up as scratches sitting next to unused run-up.
- A prolonged journal is the product. One good week is a feeling with candles attached.
Win rate is the wrong headline
A backtest that leads with win rate is selling comfort. Comfort is how often you were allowed to feel correct. The account cares about the average dollar, or the average R, after the losers. That average is expectancy. Profit factor says whether gross profit covered gross loss. Drawdown says whether you could have sat through the path that produced the profit. Those three, read together, are the account. Win rate is a detail inside them.
Put two playbooks side by side. One wins 70% of trades and pays 0.4R on a winner against a full 1R loser. The other wins 40% and pays 2R. The first feels skilled and can still bleed. The second feels ugly and can still pay. Traders abandon the second because a week of it feels like failure, then keep the first because the green days are frequent. The journal is the only referee that does not care how the week felt.
Reward-to-risk on the position tool is the plan. The journal's profit is what you collected after you moved the stop, cut the target, or scratched the trade. A planned 3R that you close at 0.5R is a different business from the one you describe to yourself. TestingView stores RR with the trade and updates it as you manage the position, so the planned ratio and the collected result can sit in the same row.
Run-up and drawdown on that row are the path, not the exit. Run-up is the best open profit the trade showed. Drawdown is the worst open loss. A winner with almost no run-up beyond the exit was a target the market barely granted. A scratch with a large run-up was a trade you left early. A loser whose drawdown blew past the original stop was a stop you moved, or a gap. Read those columns before you praise the exit.
Win rate, risk reward, expectancy, and profit factor
These four trading statistics are the ones a backtest is asked for by name. Each one answers a different question. Quoting only the win rate is how a losing playbook gets described as a good one.
| Statistic | Formula | The question |
|---|---|---|
| Win rate | Winning trades ÷ closed trades | How often did a trade close green? |
| Risk-reward ratio (RR) | Reward distance ÷ risk distance | How much did the plan aim to make per unit risked? |
| Expectancy | (Win rate × average win) − (loss rate × average loss) | What did the average trade pay? |
| Profit factor | Gross profit ÷ gross loss | Did the winners cover the losers? |
| Sharpe ratio | Average result ÷ volatility of results | Was the path smooth, or a few lucky trades? |
| Maximum drawdown | Largest peak-to-trough drop | Could you have sat through the worst stretch? |
Risk reward is the planned ratio on the position: for a long, reward is target minus entry, and risk is entry minus stop. Traders write it as RR, R:R, or 1:2. A 1:2 plan risks 1 unit to aim for 2. The break-even win rate before costs is 1 ÷ (1 + R), where R is the reward multiple.
| Planned RR | Break-even win rate |
|---|---|
| 1:1 | 50% |
| 1:1.5 | 40% |
| 1:2 | 33% |
| 1:3 | 25% |
Spread and slippage push the real break-even win rate a little higher than the table. A 1:2 plan you actually close at 1:1 is a 50% business wearing a 33% label. TestingView stores RR with the trade and updates it when you move the stop or the target, so planned RR and collected RR can sit on the same row. That gap is the statistic. The ratio you announce is not.
Expectancy uses the RR you collected, not the one you drew. Worked on R: a 40% win rate, a 2R average win, and a 1R average loss is (0.40 × 2) − (0.60 × 1) = 0.20R per trade. On a session that risks 1% per trade, that is 0.20% of the account per trade before you celebrate a win rate. Profit factor is the same sample seen as a pile: gross profit divided by gross loss. Above 1.0 the closed trades made more than they lost. A profit factor of 1.1 on a handful of trades is a coin that has not been flipped enough times. Put the spread in before you trust either number.
Sharpe ratio: was the return a smooth path?
The Sharpe ratio is the statistic people ask for when they want one number for “was the return worth the swings.” The textbook version is average excess return divided by the standard deviation of those returns: (average return − risk-free rate) ÷ volatility. Fund write-ups often annualize a daily series by multiplying by the square root of 252, and then treat a Sharpe above 1 as acceptable and above 2 as strong. Those thresholds belong to that annualized series. They do not travel unchanged onto a list of twenty trade results.
TestingView shows a simplified Sharpe ratio on the backtest: average trade result divided by the volatility of those trade results. Higher means the sample was smoother relative to what the average trade paid. You can open the Sharpe progression and watch the ratio change as trades are added. The figure does not subtract a risk-free rate, and it is not an annualized fund Sharpe. Do not line a TestingView reading of 0.8 up against a hedge-fund claim of 1.5 and call them the same fact.
Use it as a comparison inside your own journal. A chop method full of small winners can print a calmer Sharpe than a trend method that made more money in fewer, larger trades. The trend method can still be the one you want. A high Sharpe on a short sample, or on a sample that never sat through a dull month, is a smoothness score for a friendly window. Read it beside maximum drawdown, the longest losing streak, and the number of trades. If the Sharpe looks excellent and the drawdown would have stopped you trading, the Sharpe is describing a path you will not follow.
Sharpe also goes quiet about when the money was made. A smooth ratio can be an average of a paying hour and three hours that donate it back. The hour, weekday, and condition splits later in this piece are how you stop a single ratio from hiding that. Advanced statistics, including the Sharpe reading, sit on Explorer and Trader. Starter is a short look, and a short look is the wrong place to trust a volatility ratio.
The hour that actually pays
Most discretionary playbooks are secretly a session playbook. The open. The hour after a fix. The last hour before the cash close. The dead middle, where you click because the chart is open. Blend those hours into one win rate and you will fund the dead middle with the open's profits, then wonder why the month feels busy and flat.
Performance by hour is the cut that makes this obvious. One clock hour can hold the expectancy. The next three can be a donation. The fix is usually smaller than a new indicator: stop trading the hours that do not pay, or demand a stricter setup there. Traders who "trade all day" are often running two methods, one of which has no edge, and paying spread on both.
The chart timezone has to stay put for this split to mean anything. A 9:30 that is New York on Monday and exchange time on Tuesday is two different candles wearing one label. TestingView follows the chart timezone, including daylight-saving changes, and predefined-hour pauses can stop playback at the clocks you care about, such as 09:30 or 15:30, so you look on purpose instead of clicking through. Those pauses are there for the smaller replay steps. A four-hour jump will not honor a 09:30 stop, because the step never lands on 09:30.
Hour stats from a single week are a rumor. One news morning can paint 09:30 as a gold mine. The next month of ordinary opens tells you whether that hour is the method or the headline. This is why the sample has to be long enough to include dull days. The open that only works when a number hits is a different playbook, and it deserves its own tag.
There is a honesty check inside the hour, too. Analysis Time records how long you actually spent on the chart. An hour you would study for ten minutes live, and accept in two seconds on replay, is not the same decision. If the times collapse toward zero, the hour stat is a record of clicking. Slow down on the hours you claim to trade.
Day, weekday, week, and month
The day is the next lie a blended curve tells. Monday trend and Friday chop are not the same trade, even when the setup drawing looks identical. A weekday chart shows whether your playbook has a calendar. Some methods pay Tuesday through Thursday and give it back on the days when the book is thin. Some methods are a Monday-open method and nothing else. You cannot see that from a monthly profit number.
Day, week, and month views answer a different question: did the edge survive time, or did it live inside one friendly stretch? A beautiful week inside a flat month is a streak. A flat week inside a paying month might be the cost of staying in the method. Streaks belong next to these charts. A long winning run proves you can follow a rule while it works. It does not prove the rule survives the regime that comes after the run.
Equity and max drawdown sit over the top of the calendar. The question is where the hole happened. A drawdown that lives entirely inside one news week is a different risk from a drawdown that grinds for two months of ordinary conditions. Prop-firm limits care about both the depth and the clock. A strategy you cannot date is a strategy you cannot defend when the firm, or your own nerve, asks what just happened.
TestingView can show performance by weekday, hour, day, week, and month from the simulated trades in the session. Advanced statistics, including these cuts, are on Explorer and Trader. Starter will show you the workflow and a short record. It will not show you a calendar. Twenty trades cannot fill a week, let alone a month of hours.
Trades per day and per week
Frequency is a number traders treat as a personality. "I take two trades a day." "I only want A-plus setups." The journal either confirms the sentence or exposes it. Count the rows by date. Count them by week. The pace you practiced is the pace you will take live, especially once the replay speed has trained your hand.
Too many trades in a day is a cost problem before it is a psychology problem. Each fill pays spread, and often commission and slippage. A method with a small expectancy per trade can be a fine business at four trades a week and a loser at four trades an hour. Scalps die here first. The gross win rate stays flattering. The net, once each ticket pays the market to participate, does not. On the Trader plan, slippage and spread go into the statistics, so the pace is judged after the toll.
Too few trades is a sample problem. One trade a day, only on clean days, means a hundred-trade record takes most of a year. That can be the right life. It is a bad life to judge from ten trades and a feeling. Write the pace down as part of the rules: maximum trades per day, minimum minutes between them, no trades after the second loser, no trades outside the hours that pay. Then the week count becomes a compliance number. You can see the days you broke it.
Pace also changes the forecast, because a forecast is expectancy times how often you show up. Double the trades without keeping the expectancy and you have doubled the costs and the drawdown, not the edge. The week chart is where you notice a pace change. A month that suddenly has twice the trades is often a month you got bored, or a month the market got loud, and you called both of those "being active."
- Trades per day. The load on attention, and the load on spread.
- Trades per week. The real sample rate. This is how fast you will know if the month was skill.
- Trades after a loser. The revenge cluster. It hides inside a normal daily count unless you read the sequence.
- Skipped days. Days with zero trades are data. A method that cannot stand aside will invent trades on those days live.
Market conditions change the number
A setup is not a setup until you name the condition it was allowed to appear in. Trend, range, high volatility, dead volume, news, the hour after news, a level everyone can see, a level only you drew. Average those into one expectancy and the trend days will sponsor the chop days. You will feel inconsistent. The record will show two businesses, one of which you should stop funding.
The tag has to be chosen before the outcome is obvious, and it has to use the same words every week. "Breakout," "BO," and "momentum" become three playbooks in the spreadsheet and one playbook in your head. Explorer and Trader let you add a dropdown column for setup, grade, and market condition. Define the list at the start of the sample. A tag you invent after you know which days were green will group the trades the way you wish they had been grouped.
Notes and snapshots are the evidence under the tag. A dropdown says "trend." The snapshot shows whether the higher-timeframe chart agreed at the moment you clicked, or whether you promoted a chop candle after it worked. Auto-capture around entry and exit keeps the picture you traded, including the losers. The loser is the row you will want to delete. It is also the row that tells you the condition was wrong.
Conditions change inside a long sample on purpose. That is the point of a prolonged backtest. A rule that only exists inside one volatility regime should say so in the journal, and you should be allowed to stand aside when the regime leaves. Forcing the same entry through every condition is how a good trend method acquires a losing chop book. Separate sessions, on the Trader plan, keep two playbooks from sharing one curve and calling the blend an edge.
There is a forward version of the same idea. When you later trade live, the live journal should use the same condition names. The backtest says what the playbook did in a trend. The live record says whether you still take it only in a trend. Mixing the two files into one fantasy equity curve is how traders promote themselves. Keep them apart, then compare.
What a forecast is allowed to say
A forecast in a backtest is a projection of the record, not a vision of next month's price. If this expectancy, this pace, and this drawdown shape keep showing up, the next stretch of the same behavior looks like this. That is a useful picture. It tells you whether the business, repeated, is worth your time, your prop-firm fee, or a larger size. It does not tell you that next Tuesday will trend.
TestingView can show a performance forecast from the session on plans that include advanced statistics. Read it after the sample is long enough to include losers, dull hours, and at least one change in conditions. A forecast built on twelve winning trades is a compliment. A forecast built on a few hundred rows, tagged by hour and condition, is a planning tool: what the account does if you keep this job.
Use the forecast to ask operational questions. At this pace, how many weeks until the sample is no longer a story? At this drawdown, does a funded loss limit survive a normal bad month? At this expectancy per trade, what does spread do to the path if you double the number of trades? Those questions are about you. They are answerable from the journal. A price target for next quarter is a different product, and this chart will not honestly sell you one.
The dangerous way to use a forecast is to tweak the rules until the projected curve looks like a salary, then trade the tweaked version as if the projection had been collected. That is fitting, aimed at the future tense. Freeze the rules that produced the forecast. Trade those. If you change the stop, the hour, or the break-even, you owe yourself a new sample before you believe the new picture. The longer record is discussed in what a long backtest actually shows.
Optimal stop and optimal target
Traders argue about stop distance in the abstract. Wide enough to breathe. Tight enough to pay. The argument is useless until it is aimed at trades you already took. An optimal stop study asks a narrower question: on this sample, how would the results have changed if the stop had been a different distance? An optimal target study asks the same about the take-profit. TestingView can run those what-if views from the session.
The study is only as honest as the fills behind it. A stop that sits inside a spread you did not model will look brilliant and fail live. A target that assumes you were paid the high of the candle, when you would have been paid the bid, is a wish. Put slippage and spread in the statistics before you crown a distance. And remember what a single candle cannot prove: if the stop and the target both sit inside one bar, the path between them can stay ambiguous unless you played a smaller step. Optimal distances built on guessed bars are still guesses.
There is a trap in the word optimal. The distance that maximizes the past file is often the distance that memorized that file. Use the study to see the shape. Maybe your stops are so tight that noise takes you out before the move you came for. Maybe your targets are so far that run-up peaks and comes back, and you never get paid. Those are management facts. They become a new rule only after you lock the distance and collect another sample you did not use to pick it.
Manage the live distance on the chart the way you intend to manage it in the test. After entry, TestingView replaces the setup drawing with lines you can move for stop and target. If your real habit is to trail, trail in the replay, at the moment your rules say to trail. If you only tighten the stop after you have seen the rest of the move, the optimal study will describe a trader who cheats. The journal will still record it. You will feel clever. The forward month will feel confusing.
Risk percent and the stop are the same decision. A tighter stop at the same cash risk is a larger position. The optimal distance that looks best on R-multiples can be a worse distance in money once size, spread, and the day's trade count are included. Read the study in the currency of the session, next to drawdown, not only as a pretty R curve.
Break-even that fires too early
Moving a stop to entry feels like professionalism. The trade cannot hurt you. The account tells a colder story. A break-even that fires before the trade has earned the room will turn winners into scratches, raise the number of near-zero rows, and leave the real losers untouched. Win rate can rise. Expectancy can fall. You will feel safer while the method gets worse.
The signature in the journal is specific. Profit near zero. Run-up far beyond the scratch. A target that price later reached, after you were already flat. TestingView can review outcomes for trades that reached break-even, including the hypothetical path those trades still had. That review is the argument, trade by trade, against a BE trigger you placed because it felt mature.
Where the trigger sits is the rule. Just beyond costs is a different rule from one R, and one R is a different rule from "after the level breaks." On Explorer and Trader you can arm automatic break-even so the stop moves to entry when price reaches the level you set, on every trade that gets there, including the ones you would rather not re-live. Doing it by hand, only on the trades you remember, is how the sample flatters you. The line has to sit between entry and target, on the profitable side of entry.
Give the rule a long window before you trust it. A BE that looks perfect in a chop, because targets were never going to pay, will strangle a trend month. Tag the condition. Read BE outcomes inside the trend rows and inside the chop rows separately. You may keep the scratch in one condition and ban it in the other. That is a written playbook. "I go to break-even when it feels right" is how a prolonged journal dies as a source of truth.
Starter does not include automatic break-even. If break-even is part of the method you are judging, the sample needs the plan that can fire it the same way every time. Otherwise you are testing a different exit from the one you will use when the money is real.
Why the journal has to run long
Every split in this article is hungry. An hour chart with two trades in the 10:00 bucket is a coincidence. A weekday chart from one week is a calendar with no history. A forecast from a hot streak is a compliment with an axis. An optimal stop from fifteen trades is a curve fit you can do in your head. Break-even outcomes from a single news day describe that day. The numbers become something you can fund only after the same rules have been written down across a prolonged stretch: multiple weeks, dull days, at least one change in volatility, enough losers to see the drawdown, enough hours to see which clocks pay.
That is the whole job of the journal. Memory keeps the winner and misplaces the stop you widened. A screenshot folder keeps the charts you were proud of. A spreadsheet you update "later" loses the rows from the days you were tired. TestingView saves the session, the trades, the notes, the snapshots, and the settings as you go. You resume it. You do not rebuild it from a feeling on Sunday night.
Prolonged does not mean one endless mixed file. Keep the columns stable so two months can be compared. Keep the condition tags stable. Keep the timezone stable. If you change the playbook, mark the date and start a clean read, or clone the session before you experiment. Cloning copies the trades, the journal, the snapshots, and the settings, so you can compare a new management habit against the record you already trust. Editing the original until it says what you wish you had done is how a long sample becomes fiction.
A weekly rhythm is what makes the length survivable. One replay, rare and heroic, produces a story. A session you reopen on the same chart, with the same rules, produces a record. The habit is the subject of regular backtesting on TradingView. The length is what makes hour, day, pace, condition, forecast, stop, target, and break-even worth opening.
When you want the rows outside the chart, Trader exports the journal to Excel. Sort by hour. Filter the condition. Count the week. Send the file to the only person who has to believe it, which is you. The export is the simulated record. It is not a broker statement, and it should not be edited into a better past.
How TestingView gathers the record
The useful numbers are a side effect of practicing on the chart you already have open. Indicators, drawings, and the timeframes stay. You step the candles with the future hidden. You draw the position the way you would live: entry, stop, target, and size. The journal takes the row whether the trade pays or not.
Gather a sample you can actually read:
- Open the TradingView layout you trade. Start a TestingView session with a balance and a currency you think in. A fantasy balance makes every risk look tiny.
- Hide the future. Leave upcoming candles covered. Prevent Candle Previews stops a higher-timeframe switch from showing a bar that had not closed.
- Refuse your favorite week. ChronoShuffle can open the session on a random date inside a range, so you do not begin on the stretch you already remember.
- Build the journal before the first trade. Add dropdowns for setup and market condition. You will not remember the words later, and you will be tempted to improve them.
- Trade the written pace. If the rule is two trades in the morning, the replay does not get a third because you are bored. Predefined hours can pause you at the clocks that matter.
- Let break-even, the stop, and the target follow the rule at the moment the rule says so. Automatic break-even, on Explorer and Trader, applies the scratch the same way on every trade that earns it.
- Include the cost if the spread is a real part of the stop. Trader can put slippage and spread into the statistics before you read expectancy.
- Then read. Expectancy, profit factor, drawdown, and streaks first. Hour, weekday, day, week, and month next. Then the forecast, the optimal stop, the optimal target, and the break-even outcomes. A twenty-trade look cannot feed those charts. A long journal can.
Starter is the short look: one session, core stats, up to 20 trades. Explorer is where the sample starts to split: up to 50 trades, automatic break-even, advanced stats, and custom columns. Trader is the prolonged practice: no trade cap, more than one session, Excel, and cost modeling. Pick the plan that can hold the record you claim you want. See the plans, or start from the feature list if you want the charts named one by one. The knowledge base covers the controls.
The edge you can keep is a stack of boring numbers that agree with each other for a long time. The hour pays. The other hours do not. The day count matches the rule. The condition tag explains the losers. The forecast still looks like a life you want after costs. The stop and the target survive a second sample. Break-even scratches the trades it should scratch, and leaves the runners alone. TestingView is how that stack gets written down without leaving the chart.
Questions traders ask about the numbers
What is a good win rate in trading?
A good win rate is the one that clears the break-even win rate for your real risk-reward ratio, after costs. A 1:2 planned RR breaks even near a 33 percent win rate before costs. A 1:1 RR needs about 50 percent. A 70 percent win rate still loses if the average loser is much larger than the average winner. Read win rate with expectancy.
What is the risk-reward ratio (RR)?
The risk-reward ratio compares the planned reward to the planned risk. For a long, reward is target minus entry and risk is entry minus stop. A 1:2 RR risks 1 unit to aim for 2. The break-even win rate before costs is 1 divided by (1 plus the reward multiple). Realized RR is often smaller than planned RR once you exit early or pay the spread.
What is the Sharpe ratio in a backtest?
The textbook Sharpe ratio is average excess return divided by the standard deviation of those returns. TestingView shows a simplified Sharpe ratio: the average trade result divided by the volatility of the trade results. It is a smoothness score for the sample, not an annualized fund Sharpe, and a short sample can flatter it. Read it next to drawdown and the number of trades.
What is expectancy in trading?
Expectancy is the average result per trade. The formula is (win rate times average win) minus (loss rate times average loss). A 40 percent win rate with a 2R average win and a 1R average loss has expectancy of 0.20R. Positive expectancy is the edge. Win rate alone is not.
What is profit factor?
Profit factor is gross profit divided by gross loss. Above 1.0 the closed trades made more than they lost. A thin margin above 1.0 often disappears once spread and slippage are included. A very high profit factor on a handful of trades is a small sample, not proof.
Which trading statistic matters more than win rate?
Expectancy, profit factor, and drawdown decide whether the account paid. Win rate only says how often a trade closed green. A high win rate with small winners and large losers still loses money. Read those figures next to results by hour, by day, and by market condition.
Why split a backtest by hour and by day?
A playbook often pays in one part of the session and gives the money back in another. Performance by hour, weekday, day, week, and month shows which clock times fund the result. A blended win rate hides that split.
How many trades per day or per week should a backtest show?
The useful number is the count you actually take when you follow the rules, not a target you invent. Too many trades in a day raises costs and fatigue. Too few means the sample will take months. The journal timestamps are how you see your real pace.
What is a performance forecast in a backtest?
A forecast projects the record you already produced: expectancy, pace, and the equity path, if you keep trading the same way. It is a picture of the sample extended. It is not a prediction of next month's price.
How do you find an optimal stop loss and take profit?
After a real sample, a what-if study asks how the trades you already took would have finished at other stop and target distances. TestingView can show optimal stop-loss and take-profit studies from the session. Change the rule, then gather a fresh sample before you trust the new distance.
How do you know if you move to break-even too early?
Compare the scratch at entry with the run-up the trade reached, and with the outcome if the original target had been left in place. A pile of near-zero profits next to large run-up is the signature of a break-even that fired too soon. TestingView can review outcomes for trades that reached break-even.
Why keep a backtesting journal for a long time?
Hour, day, condition, stop, target, and break-even splits need many trades across more than one market mood. A single week can flatter every number. A journal kept for a prolonged stretch is the only place those splits become a record you can size from.