hypothesess.com
Ideas, put to the test. For analytical readers who want strategies tested with evidence, not opinions.
- E. coli Genome Contains 1.9 to 2.0 Bits Per Base
E. coli's genome contains approximately 1.9 to 2.0 bits of information per base, revealing its complexity and compression limits.
- Bell's test: why local recipes stop at 2 and quantum reaches 2.83
If each particle's answers are fixed in advance, Bell's test scores at most 2; quantum mechanics scores 2.83, and a few hundred pairs show the gap.
- How many trades does a backtest need before it means anything?
About 800 trades are needed for a backtest to confidently distinguish a 55% winning strategy from random chance.
- 4,096 trading strategies, 20 years of the S&P 500: none beat luck
Out of 4,096 trading strategies tested on two decades of S&P 500 data, none survived a proper correction for luck—zero proved real skill.
- How far below its peak does a 55% strategy fall? Exact numbers
Over 250 trades, a strategy that wins 55% of equal 1% bets typically falls 11% below its peak at some point, and one run in ten falls 18% or more.
- Costs versus edge: how much of a 55% strategy survives the spread
- The best of many: why your best backtest is mostly the number of tries
The more strategies you try, the better your best backtest looks, even when none of them works; the best of 100 coin-flip strategies typically shows a two-year Sharpe ratio of 1.7, then goes nowhere on new data.
- Edge or luck? How many trades before a win rate means anything
Fifty trades at a 60% win rate happens to a pure coin-flipper about one time in ten; to tell a real 55% edge from luck you need roughly 800 trades.
- Is a win due? What a losing streak does and does not tell you
After five losses in a row, a strategy that wins 55% of the time still wins its next trade 55% of the time; what the streak should change is your idea of how long losing streaks get.
- Expectimax vs Monte Carlo Tree Search for Dice Games: Which Algorithm to Choose?
Choosing the right algorithm for decision-making in dice-based games can significantly influence performance.
- Elo vs Glicko-2: Glicko-2 tracks changing strength faster
While Elo ratings offer a straightforward approach, Glicko-2's extra measures of uncertainty let it adapt faster when a player is new, returns after a break or changes strength quickly [4].
- A Winning Move Can Still Be a Bad Decision: Expected Value Versus Outcome Bias
A single victory doesn't prove you made the right move—sometimes, a winning result can come from a bad decision.
- How Deep Can Drawdowns Get for a Strategy with a 55% Win Rate?
For a strategy that wins 55% of its trades, gaining 3% of equity on a win and losing 2% on a loss, our own simulation of 1,000-trade paths puts the typical deepest drawdown near 19%, with one path in…
- Why Backtests Overfit: Multiple Testing and the Deflated Sharpe Ratio
Overfitting in backtests can create a deceptive sense of success, leading to inflated performance metrics like the Sharpe ratio that fail to hold up in live trading. This article explores how…
- Momentum vs Mean Reversion: What the Research Evidence Says
Momentum and mean reversion strategies serve different purposes in trading, and their effectiveness varies based on market conditions. Momentum strategies excel in trending markets, while mean…
- How Poker AIs Learn to Bluff Using Counterfactual Regret Minimization
A failure to bluff effectively can lead to predictable patterns that skilled opponents exploit, diminishing an AI's performance in poker. This article explores how counterfactual regret minimization…
- Kelly Criterion vs Fixed-Fraction Position Sizing: Navigating Misestimated Edges
Suppose a trader estimates a 60% win probability on a particular strategy using the Kelly criterion. They calculate their optimal bet size and aggressively invest based on this estimate.
- Why a single train/test split overstates a trading strategy
Relying on a single train/test split for evaluating trading strategies can lead to overfitting, resulting in misleading performance metrics that may not hold in real-time trading. In contrast…
- MCTS or alpha-beta: pick by whether you can score a position
Choosing the right AI algorithm can significantly impact game performance and strategy effectiveness. Monte Carlo Tree Search (MCTS) is preferable for complex, uncertain environments with many…
- Tuning backtest parameters without fooling yourself: a moving-average grid worked through
Backtesting is the process where you evaluate a trading strategy using historical data to determine its potential effectiveness. Key backtesting parameters include the time frame, the type of data…