Medium

I watched a trader spend three weeks optimizing a backtest to 87% win rate, then lose money on the first five live trades. The strategy was perfect in hindsight. It was useless in reality.

Backtesting gets misused because it's seductive. You can measure something. You can show a number. But that number answers a very specific question, and everyone treats it as answering a different one.

## What a backtest actually validates

A backtest shows you: "If I had executed this exact signal logic on this exact data, in this exact market regime, with perfect timing and zero slippage, what would have happened?"

That's useful. It tells you your logic isn't obviously broken. It weeds out strategies that lose money on *every* day in the dataset, not just most days. It lets you compare two signal approaches side-by-side under identical conditions. If one lost 40% and the other lost 8%, that difference is real information.

A backtest also constrains you to actually *define* your logic. "Buy good stocks" doesn't backtest. "Buy when RSI crosses below 30 and volume exceeds 20-day average" does. The specificity itself is the win.

But here's where most traders stop reading the fine print.

## What a backtest absolutely cannot tell you

**Market regime has changed.** Your backtest might run on three years of data. Bonds behaved one way in that period, Fed policy was one way, inflation was one way. Next quarter, one of those shifts. Your signal fires half as often. Or twice as often. The data you tested on is not a promise about the data you'll trade on.

**Slippage is way worse than you think.** A backtest usually assumes you enter at the exact close price or at the exact price your signal fires. Reality: if your signal fires at 3:58pm and you're a retail trader, you're entering at 4:00 or 4:01 at a slightly different price. Crypto? The spread is bigger. Illiquid tickers? Much bigger. A strategy that works on 2% slippage can bleed dry on 5% slippage. Your backtest never added that cost.

**You won't actually follow it.** This is the one backtests can never capture. You're down 15% on the month. Your signal fires. You pause. You "just check the news first." You second-guess. You skip that signal because it *feels* wrong even though your logic says execute. A backtest assumes you're a machine that runs the same logic every single time. You're not. Discipline under drawdown is not a number. It's a skill, and it's not included in any backtest file.

**Survivor bias hides the signals that blew up.** If you built your signal by testing thousands of variations and keeping the one that performed best, you've curve-fitted. You've selected the best-case scenario. Backtesting software lets you do this by accident. You end up with a signal that was perfect for the past but is optimized into oblivion for the future.

## How to actually use backtesting

Don't use it to predict returns. Use it to falsify your logic. Run the backtest, yes. But then ask: What would break this? What's the worst month in this data? What happens when the signal fires during a flash crash? What happens when it doesn't fire for six months?

If you can't articulate what market condition would kill your strategy, you haven't done the thinking yet.

The backtest is a **talking point, not a verdict.** It's the start of a conversation with yourself about whether your logic holds water. The real test is what happens when you deploy it for real: with real slippage, real discipline challenges, real market regimes you didn't optimize for.

That's why an approval gate matters. Every signal your AI generates, you see it, you think about it, you approve or reject. You're not trusting the backtest to be your future. You're using it as a foundation, then adding the human judgment the backtest can never include.

Start here if you want to dig deeper into backtesting without the hype: https://tradercoach.app

The Math That Finfluencers Skip

Open any trading feed and you will find screenshots of triple-digit days and the implication that consistent, spectacular returns are one course purchase away. Set that against the base rate: only 1 to 4% of day traders are profitable over the long term, and in one futures study just 1.1% cleared more than minimum wage.

Both things are not true at once. The gap between them is where most retail capital quietly goes.

The honest version is far less shareable. A real edge is usually small and shows up only over a large sample. Two traders with the identical strategy can end a year far apart purely on position sizing and whether they followed their rules on the bad days. Drawdowns are not a sign the system broke; they are a normal feature of any system with an edge, and the deep ones are mathematically brutal to climb out of.

What the screenshots never show is the denominator: how many accounts ran that same play and are not being posted. Survivorship bias is not just a backtesting flaw, it is the entire architecture of a trading feed.

Realistic expectations are not pessimism. They are the thing that keeps you sized correctly and still in the game when the person selling the dream has moved on to the next launch.

The full piece lays out what sustainable-looking numbers actually are. Read it on the blog: https://tradercoach.app/blog/

The One-Bar Mistake That Fakes a Winning Strategy

Look-ahead bias is using information in a backtest that you could not have had at the moment of the decision. It is the most flattering error in all of quantitative trading, because it lets your system trade on the future while pretending it traded on the past.

The classic case is a single bar. Your logic decides to buy based on the current candle's closing price. But in live trading you do not know that close until the candle actually closes, by which point the move you were reacting to has already happened. In a rigorous backtest, decisions must reference the previous confirmed bar, close[1], never the still-forming current bar, close[0]. One index off, and a mediocre strategy starts looking prophetic.

It hides in less obvious places too: using a day's high to time an entry, applying an indicator that quietly repaints, or filtering on data that was only reported weeks later. Each one leaks tomorrow's information into today's decision.

The discipline is to interrogate every input with one question: at the exact instant this trade fires, would I actually have known this? If the honest answer is no, the backtest is measuring a version of you that can see the future, and that trader does not show up on Monday.

The full piece covers look-ahead bias, repainting, and how to pressure-test a suspiciously good result. Read it on the blog: https://tradercoach.app/blog/

The Fastest Way to Turn a Bad Day Into a Blown Account

Revenge trading has a recognizable sequence. A loss closes. A specific urge follows: get it back, now. A new position opens with no setup, no checklist, no analysis, just the need to feel whole again. It is the single fastest mechanism by which a trader moves from a bad day to a blown account.

The mechanics underneath are worth naming. Loss aversion means a loss registers as roughly twice as painful as an equal gain feels good. That pain demands immediate relief, and the market is happy to sell you a story that the next trade provides it. What actually happens is that a defined, manageable loss gets converted into an undefined, open-ended one.

Willpower is a poor defense here, because the urge peaks exactly when your judgment is weakest. Structure works better because it is decided in advance, when you are calm. A daily loss limit that ends the session. A mandatory cooldown after a losing trade. A rule that the next position cannot be larger than the last. None of these require you to feel disciplined in the moment. They just make the destructive move harder to reach.

The full article covers the psychology and the specific circuit-breakers that interrupt the loop. Read it on the blog: https://tradercoach.app/blog/

Most Traders Don't Blow Up. They Quit.

The headline stats get quoted constantly: around 90% of day traders lose money in their first year. Less discussed is the shape of the exit. Roughly 40% quit within a single month. More than 70% are gone by six months. Only about 13% are still trading after three years.

Read that curve carefully and it tells a different story than raw failure. A large share of people never lose to the market at all. They lose to the emotional cost of an unstructured first few months, take a run of avoidable losses, and walk away before they have enough repetitions to learn anything.

That reframes the early goal. In year one, the objective is not to make money. It is to still be in the seat, with capital and composure intact, long enough for a real edge to have a chance to show up. Survival is the prerequisite for everything else.

Survival is mostly mechanical: small position sizes so no single trade matters much, a written plan you can follow on a bad day, and a journal so your losses at least buy you information. None of that is exciting. All of it keeps you on the far end of the attrition curve.

The full piece breaks down what the 13% tend to do differently in those first months. Read it on the blog: https://tradercoach.app/blog/

The Dead Companies Your Backtest Never Sees

Run a strategy over the S&P 500's current members for the last twenty years and you will likely get a flattering result. There is a problem baked into that test before you write a single line of logic: the index you are testing on is a list of winners.

Survivorship bias happens when your dataset only includes assets that made it to the end of the period. The companies that were delisted, went bankrupt, or got absorbed simply vanish from the record. Your strategy is quietly graded on a market where failure was deleted in advance.

This is not a rounding error. Research by Elton, Gruber, and Blake found survivorship bias overstated average mutual fund returns by roughly 0.9% per year. Free retail data sources make it worse, since they typically carry only active, tradable tickers and drop the bankruptcies your strategy most needs to see.

The fix is unglamorous: use point-in-time data that includes the names that later died, and ask of any backtest, what is missing from this dataset, not just what is in it.

The rest of this piece walks through survivorship bias, look-ahead bias, and the other traps that make a backtest look better than the strategy actually is. Full write-up on the blog: https://tradercoach.app/blog/