Why Most Traders Lose Money — And What the Data Shows

The market did the work

Almost every argument about trading eventually becomes an argument about ability. One side believes that reading charts, timing entries, and cutting losses can be learned; the other believes it cannot. Both sides tend to skip the step that should come first: asking what the market returned during the period in question, and what it costs to try to capture that return.

That first number is easy to establish. Between 3 April 2017 and 4 September 2026, across 2,363 daily closes, the S&P 500 futures contract gained 227.8%. Annualized, that equals 13.5% per year. Capturing this return required one decision at the beginning and no further decisions afterwards.

That figure is a floor rather than a headline. It is a price series without dividends, and two independent checks put it below the market it represents: compared with the ETF tracking the same index, it is 0.7 percentage points lower per year; compared with the NAV performance reported by an S&P 500 ETF in our own fund data, the gap is 1.2 percentage points per year. Every comparison that follows therefore gives the trader the benefit of the doubt.

So the return available from the market was substantial, and any explanation for poor trading results has to show where a return of that size disappears. Five forces account for most of the difference. None of them requires the trader to be wrong about the market, and all five can be measured using our own price data. Before looking at the first one, however, it is worth defining what type of trading we are discussing, because the answer changes the size of every number that follows.

Which trading this is about

Day trading, swing trading, and position trading differ in many ways, but only one factor can be measured directly here, and it is the one that matters: how long a position is held. Everything else — costs, the number of round trips, and the size of a typical gain — follows from it.

Our database contains daily closing prices, so holding periods from one session to a full quarter can be measured. Anything that happens within a single day cannot. Intraday trading is therefore outside the scope of this study, and nothing below should be read as a statement about it.

For each holding period, the market’s own history answers three questions at once: how often a long position was profitable at the end of the window, how large the average gain and average loss were, and therefore what hit rate a trader would need to break even after paying 0.2% in costs per round trip.

The pattern is clear. At a one-day horizon, the required hit rate of 64.1% is 9.6 percentage points higher than the 54.5% delivered by the market itself. A trader therefore starts the day-to-day game with a gap that skill must first overcome before the strategy can break even. Extend the holding period to one week, and the relationship reverses: the required rate of 58.5% is below the 60.8% that simply being invested produced. At a quarter, the gap has grown to 25 percentage points in the trader’s favor, while the annual cost burden has fallen from 39.6% to 0.8%.

Two consequences follow. First, the shorter the trading style, the more skill is not a bonus but a precondition. Second, when this study talks about traders losing money, it is mainly referring to the faster end of that scale — daily and weekly holding periods, where the hurdle is highest. Position trading over months and quarters faces the same arithmetic, but much more gently.

Losses cost more than they look

The first force is arithmetic, and it is one of the most consistently underestimated. Percentages do not work symmetrically in both directions: a position that falls by 20% needs a 25% gain to return to its starting point; one that falls by half needs to double; and after a 70% decline, it takes a 233% gain simply to get back to where it started. Every further step down widens the gap between what was lost and what is needed to recover it.

The spring of 2020 shows what this means in practice. The market fell 34.4% between 19 February and 23 March. To return to its previous level, it needed a gain of 52.5%, which it reached on 21 August, 184 days after the peak. An investor who held through the decline recovered the full loss. An investor who sold at the bottom locked in the loss and left the recovery to someone else.

Individual stocks show the same mechanism in a more extreme form. Of the 545 companies in our sample, 40.4% currently trade more than 20% below their own two-year high, and 4.6% are down by more than half. The median company would need a gain of 18.6% just to return to a level it has already reached. The true figures are likely worse: the sample cannot include companies that failed, and 23 additional series had to be removed because unadjusted stock splits cannot be distinguished from crashes in a price series. This removes genuine crashes along with the data errors.

This is why risk control is not simply a question of temperament. Accepting a deep drawdown is not brave; it means taking on a recovery task whose cost rises faster than the loss that created it.

The return lives in a handful of days

If the market pays well over nine years, the obvious idea is to be invested during the good periods and out of the market during the bad ones. The data gives a clear answer to that idea, starting with how unevenly returns are distributed over time.

Of the 2,362 trading days in the series, remove the ten strongest and the result falls from +227.8% to +77.7%. Remove the twenty strongest and only +29.0% remains. Remove the thirty strongest, which represent 1.3% of all sessions, and nine years in a rising market end at −1.4% — effectively nothing.

The mirror image is just as dramatic and deserves attention because it is the strongest argument for market timing. Without the ten worst days, the same period would have returned +531.4%. If the two groups could be separated, timing would be the most valuable skill in finance. They cannot be separated, and that is the finding.

Fourteen of the twenty best days occurred within ten trading days of one of the twenty worst days. They cluster in the same few weeks — 13 and 24 March 2020, 6 April 2020, and 9 April 2025 — because sharp recoveries are part of sharp declines. They are not separate events that happen later, when markets are calmer.

The practical consequence is uncomfortable. Selling after a sharp fall is one of the most natural reactions there is, but it is also one of the most reliable ways to be out of the market when it rebounds. Avoiding the worst days while staying invested for the best ones is not a difficult skill. It is a contradiction.

A high hit rate proves less than it seems

Traders who keep records usually track their hit rate: the share of positions that end in profit. It is an intuitive measure, but it contains far less information than its popularity suggests.

Over the nine years, 54.5% of all sessions closed higher than the previous day, so the market rose more often than it fell. The average winning day, however, gained 0.75%, while the average losing day lost 0.78%. The ratio between the two, known as the payoff ratio, was 0.96: gains were slightly smaller than losses.

Direction alone therefore produces no return. What produces the return is the size of individual moves, and those moves are distributed in a way that most trading rules quietly ignore.

Days more than four standard deviations from the mean make up 0.64% of the series. Under a normal distribution, they would account for roughly 0.01%, so they occur about sixty times more often than a normal model would suggest. The measured kurtosis is 17.0, compared with 3 for a normal distribution, and the skewness of −0.29 shows that the extremes are slightly tilted to the downside. A position size that survives ninety-nine days out of a hundred is not conservative when the hundredth day carries this much risk.

The same pattern appears when the question moves from days to companies. Across the 545 stocks, the correlation between the share of winning days and total return is 0.40, while the rank correlation is 0.62. The two measures therefore move together, which is not surprising: a stock that rose over two years necessarily had more up days than down days. What the numbers do not support is the reverse interpretation — that a high hit rate produced the result. The gap between the two measures shows why. The rankings agree relatively well, while the linear relationship is much weaker. This is what happens when the largest gains come from companies whose hit rate was unremarkable. The hit rate truly fails as a guide at the level of the market itself, where a payoff ratio of 0.96 means that being right more often than wrong produced no return.

The same sample also answers a question most people ask sooner or later: whether picking individual stocks is worth the effort. The median company returned 20.3% over its period. Compared with the index over exactly the same window, only 34.5% of the companies beat it, while the average return of +49% is lifted by a small group of extreme winners. Stock picking is therefore not a coin flip with even odds. It is a draw from a distribution in which the typical outcome is far below the average, and missing the few outliers is the normal outcome rather than bad luck.

Costs decide what is left

So far, nothing has been traded. Now assume a trader who neither adds skill nor destroys value: they earn the same gross return of 227.8%, but collect it through a certain number of round trips per year, each costing 0.2% of the position.

Five round trips a month sounds modest, and it is: one position closed and reopened each week. At that pace, nine years of a rising market leave +6.3%. Over the same 111 months, German overnight deposits paid 1.76% in total. The index figures are in dollars and the deposit rate is in euros, so a fair comparison requires conversion: in euro terms, the same index period returned +200.7%. The trader therefore stays ahead of a savings account but keeps only about one thirty-sixth of what the market provided for free.

The 0.2% in the table is the cheapest case, not an average. It comes from the lowest fee in our own broker comparison. The other levels in the figure come from the same source. Trade the same strategy in €500 orders and the cost doubles to 0.4% per round trip, turning +227.8% into −65.6% at five trades a month. Trade with a direct bank charging €4.90 plus 0.25% per order, and a €1,000 round trip costs 1.5%. At that rate, even one trade a month is enough to turn +227.8% into −40.1%.

Costs are therefore not a detail to optimize once the strategy works. At ordinary trading frequencies, they are the dominant factor, and the choice of broker can change the result more than most changes to the strategy itself.

Without an edge, speed only shortens the story

The calculation above still gives the trader the market’s return. The harder question is what happens when the direction of each trade is simply right or wrong at random — the honest starting assumption for anyone who has not yet shown otherwise.

To answer it, we simulated 30,000 accounts. Each trade uses a real daily move from the index series, assigns its direction by coin flip, and pays the same 0.2% in costs. There is no skill in the model, but there is no incompetence either. After one year, 25.0% of the accounts trading once a month are ahead. At four trades a month, the figure falls to 10.4%; at ten trades, to 2.5%; and at twenty trades, to 0.3%. Extend the horizon to five years at ten trades a month, and the share of winning accounts rounds to zero, while the median account is down 71.1%.

Drawing single days at random removes something real, because quiet days tend to follow quiet days and violent days tend to cluster. Repeating the whole exercise with five-day blocks, which preserves this clustering, moves the one-year figure at four trades a month from 10.4% to 8.9%. The direction of the result therefore does not depend on the assumption; if anything, the simpler assumption is the more favorable one.

It is worth being precise about what has happened here. Nobody in this model is punished for a bad forecast, because the forecasts are fair by construction. Frequency alone turns an even game into a near-certain loss, because every trade pays a toll and those costs compound just as returns do.

This also answers a question that comes up whenever short-term results are disappointing: whether things improve with time. For an account without an edge, they do not, and the reason is the same arithmetic seen from a different angle. At four round trips a month, 10.4% of accounts are in profit after one year. After two years, it is 3.8%; after three, 1.7%; after five, 0.3%; and after ten years, the share rounds to zero, while the median account has lost 62.9%. Time does not average out the friction. It multiplies it, and the small chance of being ahead shrinks with every year that the account keeps trading.

That is the honest answer to the question of experience, too. Our data contains prices, not people: we cannot observe how long an individual has been trading, and this study makes no claim about it. What the data does establish is the size of the hurdle that experience would have to clear — 9.6 percentage points of hit rate at a daily holding period, or 12.8 percentage points per year at five round trips a month. Anything that does not clear it is not slow progress. It is a different outcome from the one shown in these charts only by luck.

Leverage works differently, and the difference is not in the trader’s favor.

Leverage puts a deadline on the trade

At double leverage, 0.1% of the accounts are forced to close out within the year. At five times leverage, it is 17.4%; at ten times, 56.3%; and at the thirty times leverage that all seven CFD providers in our broker comparison still offer, 94.8% of accounts are closed out within twelve months.

The second bar in the figure matters as much as the first. Running the same accounts with no costs at all changes almost nothing: 14.8% instead of 17.4% at five times leverage, and 93.7% instead of 94.8% at thirty times. Unlike the frequency result, this is not about fees. It is the market’s own volatility reaching a position that is too large to survive it. Repeating the exercise with five-day blocks, which keeps calm and violent periods together, gives almost the same result again: 15.6% and 94.6%.

Leverage therefore does not make anyone more likely to be right. It shortens the time until an ordinary run of bad luck becomes irreversible, and it does so at the worst possible moment: a closed-out account cannot take part in the recovery, which is precisely when the largest single-day gains arrive. The first section then applies in full, except that there is no position left to recover.

What the numbers settle, and what they leave open

Taken together, the five effects point in the same direction. A loss requires a larger gain to undo it. The return of a whole decade is concentrated in days that cannot be identified in advance and that cluster within the worst weeks. A good hit rate does not carry a portfolio, and neither does picking the typical stock. Ordinary costs at ordinary trading frequencies can consume an entire bull market, while leverage puts a deadline on the whole exercise.

None of this proves that profitable trading is impossible. It measures the size of the hurdle before ability is even considered: at five round trips a month, a trader has to beat the market by about 12.8 percentage points a year merely to match someone who did nothing.

Two factors would make that hurdle higher, and both were left out of the calculations. Financing costs on a leveraged position are not modeled, although the US policy rate stood at 3.63% in early September 2026. Taxes on realized gains are also not modeled, and these penalize frequent trading more than simply holding. Adding either factor would move every result in the same direction.

Five limitations should be kept in plain sight. Nothing here covers intraday trading, because the database contains daily closing prices and the shortest measurable holding period is one session. Nothing here measures individual experience either, since the data contains prices rather than trading accounts; the study can only state the hurdle, not who clears it. The stock sample contains only companies we currently track, so failed companies are missing, and the 23 series removed for unadjusted corporate actions also remove genuine crashes from the sample. The simulations describe a trader without skill, which is a benchmark rather than a portrait of anyone in particular. And all of it rests on a single stretch of history that happened to contain one crash and one long bull market. Another nine years would produce different magnitudes, even if the arithmetic of costs and recoveries remained the same.

The most useful finding is also the least exciting. Over these nine years, the decision that produced the best result was the decision to stop making them.