The words strategy sellers use, defined plainly — and what we actually found when we tested for each one across 1,000+ strategies and 1,700+ indicators. No jargon for its own sake. Every term links to the finding behind it where we have one.
How a strategy fails our audit
No real edge (no gross edge)
A strategy has an edge when its rules beat a random coin-flip entry on the same bars, before a cent of cost comes off. No gross edge means the signal never predicted anything. The entries were noise dressed as logic. This is the single most common reason we reject a strategy, and it's the finding people least expect. Nearly half of everything we throw out failed right here, before slippage, spread, or commission entered the math. The popular story is that costs kill strategies. Mostly they don't get the chance. Most rejected systems were never alive. When you strip the equity curve back to whether the rule actually forecast the next move, the answer is usually no.
The strategy made money because the market drifted up while it happened to be long. That's beta, market exposure, dressed as skill. Buy-and-hold would have done as well or better, with fewer rules to break. It shows up constantly in long-only systems backtested over a bull decade. In our audit trend-beta is the second most common reason we reject a strategy, behind only having no edge at all. The test is blunt: compare the system to simply holding the instrument over the same window. If a passive position matches the curve, the rules added nothing. A rising market flatters almost any long strategy. Our job is to strip the drift out and see what, if anything, is left.
Nearly all the profit came from a tiny number of outlier days. Strip out the best handful and the curve flattens or turns negative. The strategy didn't have an edge across time; it caught a few jackpots and coasted. We test this by removing the top slice of winning days and re-running. A real edge survives losing its best 10 percent. A lottery ticket doesn't. Tail-concentrated profit is our third most common reject reason, and it's the most seductive, because the headline return looks spectacular. The problem is you can't schedule those days, and live you'll usually miss the biggest ones to a gap, an outage, or a bad fill. A curve resting on five green candles is not a system.
The strategy had a genuine edge on paper, then real spread, commission, and slippage ate it whole. Gross-positive, net-negative. This is the failure everyone expects to be the big one. It isn't. Cost-fatal is only our fourth most common reject reason. The louder story, that trading costs quietly kill good systems, is real but oversold. Far more strategies had no edge to lose in the first place. When costs are the killer, timeframe is usually why: the shorter the bar, the more the spread matters. Sub-15-minute FX dies here so reliably that we no longer test FX or CFD strategies below the 30-minute chart. A system can be right about direction and still lose every month to its own transaction costs.
Tune enough parameters and any strategy will fit the history perfectly, including the random noise that won't repeat. That's curve-fitting: a system optimized to a past that's already gone. It looks immaculate in-sample and falls apart on data it never saw. The tell is fragility. Nudge one setting and the beautiful curve craters. It memorized the history; it didn't learn a rule. We guard against it by testing out-of-sample, on years and instruments the optimizer never touched. Pure overfit is a smaller reject bucket than you'd guess, because most overfit systems trip our earlier checks first, having no real edge to begin with, just a well-fitted mirage. A strategy with fourteen tuned inputs hasn't found an edge. It's remembering.
The backtest uses information that wouldn't have existed yet at the moment of the trade. A signal computed on the day's close but acted on at the day's open. A repainting indicator that quietly revises past bars. The result is a curve that could never have been traded. Look-ahead is the most dangerous bug in backtesting because it produces the most convincing fake. In our pipeline every strategy that survives the numbers gets a second pass where our strongest model adversarially hunts for exactly this: a fill that peeked, a value that leaked from the future. It's a smaller slice of our rejects than no-edge, but a costlier one, because these are the systems that look best right up until you trade them and they fall apart.
Shuffle the data, randomize the trade order, or scramble the signal, then see if the edge survives. If a strategy performs just as well on noise as on the real series, its results were luck. This is the placebo control that most published backtests skip entirely. We run it on everything. A meaningful share of systems that looked profitable collapse the moment their timing is randomized, because the pattern lived in the fit, not the market. Permutation testing is unglamorous and it makes a lot of pretty curves disappear, which is exactly why it belongs in the pipeline. If your strategy can't beat a shuffled version of itself, you found a coincidence and named it a system. Cheap to run, brutal to fail.
When a bar covers a range of prices, the backtester has to decide the order in which the high, low, open, and close happened inside it. Get that order wrong and a trailing stop or intrabar exit gets filled at a price that never would have triggered in that sequence. This is where a lot of trailing-stop systems quietly cheat. The naive bar-by-bar assumption hands them exits they couldn't have caught live. We measured the gap: bar-by-bar fills overstate trailing-stop strategies by 42 to 84 percent versus a realistic intrabar model. That's not a rounding error. For a whole class of systems that overstatement is the entire product being sold. The edge lives inside an assumption about the inside of a candle, and it doesn't survive contact with real ticks.
Double your position after every loss so the next win recovers everything, then start again. On paper the account only ever grinds upward, because you refuse to book a loss. In reality each losing streak grows the position geometrically until one run empties the account. Martingale is the engine underneath most grid and DCA bots, whether or not they use the word. It's why they show near-perfect win rates and then die in a single trend. We tested 76 grid, DCA, and martingale systems across futures, FX, and crypto. Every one failed, a 100 percent rejection rate, the only category in the whole audit to score it. The math isn't a tuning problem you can fix with a wider grid. It's the design.
A grid places a ladder of buy and sell orders at fixed intervals, aiming to harvest the market's back-and-forth chop. In a sideways range it prints steady small wins and looks unstoppable. Then the market trends, the losing side of the ladder stacks up, and open drawdown swallows every profit the grid ever booked. We ran 76 grid and DCA systems through real bid/ask across three asset classes. All of them failed, 100 percent, no survivors, the sharpest single result in the audit. The failure isn't bad settings. It's structural: a grid is short volatility and long hope, and eventually volatility sends the bill. The tidy equity curve is real right up to the day it isn't.
In investing, dollar-cost averaging means buying a fixed amount on a schedule, a sensible and boring habit. In the trading-bot world the term got hijacked. Here DCA means averaging down into a position that's already losing, buying more as it falls to pull your entry lower. That's martingale wearing a respectable name. The bot shows a high win rate because it holds and adds until price eventually ticks back, closing green while the open loss quietly balloons underneath. We tested these alongside grids: 76 systems, every one failed after real costs. Averaging into a loser works until the loser keeps going, and the market doesn't owe your average price a bounce. The strategy that always wins is usually the one hiding its losses in trades it won't close.
Win rate is the share of trades that close green. It's the first stat every strategy seller quotes and the most misleading one they could pick. A system can win 95 percent of the time and still be a guaranteed loser, if the rare losses are large enough to swallow all the small wins. That gap is the whole martingale and grid trick: by averaging down and refusing to close red, a bot manufactures a gorgeous win rate while one bad trend erases the account. High win rate is a design choice, not an edge. We care about what the winners make against what the losers cost, across the open drawdown, not the percentage of green tickets. Ask any bot advertising a 99 percent win rate what its worst single loss looks like.
Drawdown is the drop from an equity peak to the following trough, how far underwater you go before a new high, measured in percent or currency. It's the number that tells you whether you could actually have held the system without blowing up or bailing out. Closed-trade drawdown is honest. Open-position drawdown is where grid and DCA bots hide the body. Because they refuse to close a loser, their booked equity looks smooth while the real, unrealized loss balloons in positions still open on the book. The account can be down catastrophically while the trade log shows an unbroken run of wins. We measure the open drawdown, not the flattering closed curve. A win rate near 100 percent and a drawdown that can end the account are the same strategy described two ways.
Slippage is the gap between the price you expected and the price you actually got. Markets move between your decision and your fill, and the difference comes out of your account, trade after trade. Most backtests assume you're filled at the exact signal price, a gift no real broker gives you. Model it honestly and thin, fast, or short-timeframe systems lose a chunk of edge they never had to spare. Slippage is one of the three costs, with spread and commission, behind our cost-fatal rejects: the systems that were gross-positive and net-negative. It hits hardest where trades are frequent and size is large relative to what's resting on the book. A fill you assumed is free is a number you borrowed from a curve that won't pay it back live.
The spread is the gap between the best bid and the best ask. You buy at the higher price and sell at the lower, and the difference is a cost you pay on every round trip, win or lose. It sounds trivial until you cross it thousands of times. We model it from real tick data with actual bid/ask quotes, not a flat assumption bolted onto close prices. The finer the timeframe, the more it bites. Sub-15-minute FX dies almost entirely on spread, the edge real but smaller than the cost of harvesting it, which is why we stopped testing FX and CFD strategies below the 30-minute chart. A backtest run on midprice quietly deletes this cost. Real accounts can't.
Tick data is every individual quote and trade, with the real bid and ask at each moment, not a single price sampled once per bar. It's the raw material honest cost modeling actually needs. Backtest on tidy candle closes and you never see the spread you'd really pay or the fill you'd really get. For FX we use tick data with genuine bid/ask so spread and slippage are measured, not guessed. For futures we run 13 years of CME data. The difference isn't academic. A system that clears on candle midprice can be underwater the moment you charge it the real cost of every entry and exit. Cheap data gives cheap answers. If the cost model is a guess, so is the verdict on the strategy.
Walk-forward testing tunes a strategy on one slice of history, then tests it on the next slice it has never seen, and rolls that window forward through time. Out-of-sample is the whole point: if an edge is real, it works on data the optimizer couldn't peek at. If it only shines on the years it was fitted to, it memorized noise. This is the check that separates a genuine rule from a well-decorated coincidence, and it's the one most published backtests skip because it makes results look worse. We test on years and instruments held out of any tuning. A strategy that needs to see the answer before the test isn't forecasting. Real edges are boring that way. They keep working on data nobody fitted them to.
Futures contracts expire, so to backtest across years you stitch the expiring contract onto the next one into a single continuous series. Do it carelessly and you introduce phantom gaps or profits at every roll that no trader ever captured. Back-adjusting, shifting the old data so the seam is smooth, fixes the chart but can push older prices negative or distort percentage returns if you're sloppy about method. It's a quiet data trap: the strategy looks like it earned money that was really just an artifact of how the series was glued together. We run 13 years of CME futures and handle the roll deliberately, because a bad stitch can manufacture an edge out of nothing. The contract you backtest has to match the contract you could actually have traded.
Intermarket analysis claims markets talk to each other in sequence — bonds move first, stocks follow, the dollar leads commodities. Spot the early mover and you've got a forecast for the late one, or so the theory goes. The logic is intuitive enough that it became a fixture on trading desks. Interest-rate markets are famously quick to price new information, and a spike in credit stress tends to hit equities before it fully shows up in economic data. Chain enough of these relationships together and you get a story where one market's move today writes tomorrow's headline in another. The appeal is obvious. If bonds really lead stocks, you don't need a crystal ball. You need a calendar.
Real yields are the interest rate left after expected inflation, and for a long stretch they moved gold in near lock-step reverse. That relationship just broke. Real yields are the interest rate left after you strip out expected inflation — the nominal yield on a bond minus the market's breakeven inflation estimate for the same maturity. It's the actual return a lender locks in once rising prices get taken out of the math, and it's the number that decides whether holding cash-like assets pays or merely keeps pace.
Pre-FOMC drift is the tendency for equity indices to climb in the hours before a scheduled Fed decision, before any news actually breaks. Pre-FOMC drift is the tendency for broad equity indices to climb in the hours before a scheduled Federal Reserve rate decision, regardless of what the Fed actually announces. It happens before the news breaks. The pattern was first documented in the academic literature studying decades of FOMC meetings, concentrated almost entirely inside that narrow pre-announcement window.
Implied volatility is usually a lie — a flattering one, paid for by whoever is short the crash. The volatility risk premium is that lie, monetized. Options almost always price in more volatility than the market ends up delivering. Implied volatility exceeds realized volatility in a large majority of months, and that persistent gap is the volatility risk premium (VRP). Sell volatility (options, variance swaps, a strip of straddles) and over time you're paid to be short that gap.
A regime tells you how big the next move might be. It almost never tells you which way it goes. Every return carries two separate pieces of information, and traders routinely confuse them. The first moment is the mean — direction, whether price went up or down and by how much on average. The second moment is the variance — how big the swings were, and how they moved together with other assets. One tells you where. The other tells you how hard.
Every option has a market-maker on the other side of it, and market-makers hedge. Dealer gamma, also called gamma exposure or GEX, is the aggregate sign of that hedging flow across the whole options book, and it decides whether the tape grinds or gaps. When dealers are net long gamma, meaning they've sold more optionality than they've bought and the market is mostly buying protection, a rally forces them to sell into strength, a drop forces them to buy into weakness, and that hedging flow leans against whatever direction started it, compressing range and producing the grinding, low-drama sessions traders call pinned. Flip the sign and the mechanism reverses. Short-gamma dealers sell into falling prices and buy into rising ones, adding force to whichever direction already has the ball. Same market structure, opposite consequence.
The Sharpe ratio is the number traders reach for to sound rigorous — return per unit of risk, in one clean figure. It's also one of the easiest numbers in trading to flatter. The Sharpe ratio measures return per unit of risk: take a strategy's average return, subtract the risk-free rate, divide by the volatility of those returns. A strategy earning 20% a year with wild swings can score worse than one earning 12% smoothly. The metric exists specifically to price that volatility.
Multiple testing is the reason a good backtest is the least surprising result in the audit. Test enough strategies and some will look profitable by chance alone; the question is never whether you found one, it's whether the bar you cleared was high enough to matter. Test one strategy and a win rate a shade above a coin flip looks like an edge. Test a thousand strategies and, by chance alone, several will clear that same bar with room to spare. None of them found anything real. Someone just had to win the lottery. Multiple testing (also called data-snooping) is what happens when that number of attempts gets left out of the significance math. The usual pass/fail line for a backtest, a t-statistic near 2, was built for testing one idea at a time. Run a thousand ideas through that same line and the false positives outnumber the real signals before anyone has even started looking.
A repainting indicator shows you a different history than the one it displayed live. The signal you traded off has usually already moved by the time you look back at it. A repainting indicator shows you one thing on the chart today and something else on the same bar tomorrow, with no code changed in between. The arrow that flagged a perfect bottom on Tuesday's close can slide or vanish by Thursday, because the calculation behind it was never finished when it first appeared. What you saw in real time and what the history shows now are two different records of the same candle.
Profit factor is gross profit divided by gross loss, and it's one of the easiest backtest numbers to inflate without changing a single trading rule. Profit factor is gross profit divided by gross loss. Add up every winning trade, add up every losing trade, divide the first by the second. Cross 1.0 and the strategy made more than it lost, at least on paper. It's one number, easy to compute, easy to slap on a sales page.
Before you go: grab FoxAlgo Data Studio free — real tick-level market data, auto-aggregated to any timeframe, export-ready for NinjaTrader, MT4/MT5 and CSV.
Every strategy we tested — named, with the verdict
The full audit: 1,000+ strategies and 1,700+ indicators, each with its verdict and the exact reason it lived or died. Downloadable lists — the winners, the failures, and the method.