Asymmetry
The average win must be large enough to compensate for a low hit rate. Breakeven falls as realized payoff ratio rises.
p* = L / (W + L)
Probability, not prediction
You can be wrong most of the time and still have an edge.
A tail-edge system accepts frequent, controlled losses in exchange for rare payoffs that are several times larger. The win rate looks bad. The distribution is uncomfortable. The arithmetic can still be positive.
Outcome distribution100-trade illustration
70%Win rate? No. Loss rate.
+20RNet over 100 trades
5 tradesCreate +40R gross
Lose smallDefine risk before the outcome is known.
Lose oftenTreat failed attempts as the cost of access.
Win largeDo not truncate the right side of the distribution.
SurviveSize for the streak, not for your confidence.
01 / Expectancy
The mean matters more than how often you are right
A strategy is not defined by its win rate. It is defined by every possible outcome, its probability, and its size. Expected value combines them in one number.
Expected R per trade
E[R] = pW (1 − p)L
Let p be the probability of a win, W the average winning R-multiple, and L the absolute size of the average losing R-multiple. Positive expectancy means the probability-weighted gains exceed the probability-weighted losses.
RInitial planned risk. It normalizes unlike trades.pObserved win probability, not a feeling of confidence.W / LNet average outcomes, including costs and slippage.Outcome bucketCountTotal
Full losses at −1R70−70R
Ordinary wins at +2R25+50R
Tail wins at +8R5+40R
Net / expectancy100+20R
The tail's contribution
+0.20R
The average win is 3R, so 0.30 × 3R − 0.70 × 1R = +0.20R. Remove only the five +8R outcomes and the same sequence finishes at −20R. The edge is concentrated in the tail.
Illustrative arithmetic, not a promised return. Real outcomes are not limited to three buckets and may include losses beyond −1R.
02 / Distribution
Four pieces have to work together
A positive average is necessary, but it is not enough. The path, dispersion, sizing, and reliability of the estimate determine whether the edge is tradable.
Dispersion
When a few outcomes carry the result, variance is high. A positive system can look broken for a surprisingly long sample.
SE = σ / √nSequence risk
The order of wins and losses does not change arithmetic expectancy, but it changes drawdown, behavior, and survival.
P(loss) = 1 − pPosition size
A favorable distribution can still be ruined by excessive exposure. Geometric growth punishes large percentage losses.
g = E[ln(1 + fR)]
03 / Work the numbers
Test the payoff, not the story
Change the observed win rate and average outcomes. The calculator works in R-multiples, before any claim about how much capital should be risked.
04 / The path
A ten-loss streak is not evidence of failure
With a 30% win rate, losses occur with probability 70%. The chance of at least one run of ten consecutive losses in 100 independent trades is about 58%. Painful does not mean improbable.
| Consecutive losses | Chance in 100 trades | Equity after streak at 1% risk | At 2% risk | At 5% risk |
|---|---|---|---|---|
| 5 | 99.9% | 95.1% | 90.4% | 77.4% |
| 8 | 85.8% | 92.3% | 85.1% | 66.3% |
| 10 | 58.0% | 90.4% | 81.7% | 59.9% |
| 12 | 32.9% | 88.6% | 78.5% | 54.0% |
| 15 | 12.1% | 86.0% | 73.9% | 46.3% |
Run probabilities use an exact recursion for 100 independent Bernoulli trials with p(win) = 30%. Equity columns assume fixed-fraction sizing, exact −1R losses, and no gaps or costs.
05 / Necessary skepticism
Asymmetry is a shape. It is not automatically an edge.
The right tail matters only if there is a repeatable reason for it to exist after implementation. Three common shortcuts turn good arithmetic into bad inference.
A 5:1 target does not mean a 5R average win
Most trades may reverse before the target, and management rules may cut winners early. Use realized net outcomes, not the labels on an order ticket.
A stop does not guarantee a −1R loss
Gaps, slippage, liquidity, and correlated exits can push losses past the modeled boundary. The left tail exists too.
A backtest mean is not the true mean
Small samples, selection, overfitting, regime changes, and a handful of outliers can inflate the estimate. Put uncertainty around every edge.
The notebook
Read one variable at a time
Detailed guides to the equations, assumptions, and practical consequences behind asymmetric systems.
How a 30% win rate can produce positive expectancy
Derive the breakeven hit rate, profit factor, variance, and sample-size problem from one clean example.
Read the derivation →The losing streak your backtest owes you
Why low hit rates create long runs of losses, and how to distinguish an expected streak from evidence of decay.
Study the probabilities →Skew, fat tails, and where the return really comes from
Mean versus median, positive skew, outlier dependence, and why removing the best trades can erase the result.
Follow the right tail →
The source shelf
Books for thinking in distributions
Primary reading on randomness, tail risk, expectancy, trend following, and the discipline required to hold an asymmetric payoff.
Fooled by RandomnessNassim Nicholas Taleb
Start here / Randomness
Why outcomes hide process, and why a lucky result can look exactly like skill.
View on Amazon →The Black SwanNassim Nicholas Taleb
Tail events
The modern argument for taking rare, consequential observations seriously.
View on Amazon →Trade Your Way to Financial FreedomVan K. Tharp
Expectancy / Sizing
The practical vocabulary of R-multiples, expectancy, and position sizing.
View on Amazon →Following the TrendAndreas F. Clenow
Applied asymmetry
A clear, quantitative look at a strategy built around many small misses and a few large trends.
View on Amazon →As an Amazon Associate, TheEdgeIsTheTail.com earns from qualifying purchases.
Questions
Frequently asked
Can a system be profitable with a 30% win rate? +
Yes, if average wins are large enough relative to average losses after costs. At a 30% win rate, a 3R average win and 1R average loss produce 0.30 × 3 − 0.70 × 1 = +0.20R per trade. That is a model average, not a promise about the next sequence.
What is an R-multiple? +
R is initial planned trade risk. If the planned loss is $100, then losing $100 is −1R and gaining $300 is +3R. R-multiples normalize trades across account sizes and instruments. A gap can still make the realized loss worse than −1R.
Does high reward-to-risk create an edge? +
No. A proposed 5:1 payoff says nothing about how often +5R will actually occur. Edge requires a positive net outcome distribution. Win probability, partial exits, slippage, costs, gaps, and missed trades all belong in that distribution.
How do I know whether the edge is real? +
You never know with certainty. Demand a plausible mechanism, out-of-sample evidence, enough independent observations, stable results across reasonable parameters, realistic costs, and an explicit confidence interval. Then size for the possibility that the estimate is wrong.
Is this financial advice? +
No. This site is educational. Trading involves substantial risk of loss; execution can be worse than modeled; and historical, simulated, or hypothetical performance does not guarantee future results. Consult a qualified professional before making financial decisions.