What is Fitness?
Fitness in practical quant workflows
What is Fitness?
In genetic algorithms, fitness describes how well an individual performs in its environment.
In darwintIQ, the individuals are trading models — and the environment is the current market.
Fitness therefore measures how well a trading model behaves right now, under recent market conditions. It reflects not only how much a model earns, but how it earns: whether its returns are stable or erratic, smooth or fragile, consistent or lucky.
Two models can produce the same profit, yet have very different fitness. One may trade steadily with controlled risk, while the other depends on a few unstable spikes. Fitness captures this behavioral difference.
Why Fitness matters in evolving models
In evolutionary systems, selection is driven by fitness.
Models with higher fitness survive, reproduce, and shape the next generation.
This is fundamentally different from traditional backtesting, where a single strategy is optimized once on historical data. In darwintIQ, many models continuously compete inside a moving market window. The goal is not to find the most profitable model ever observed, but the models that are currently well-adapted.
A model with high profit but unstable behavior often loses fitness quickly when conditions shift. A model with coherent, stable structure tends to maintain fitness longer. Evolution naturally favors the latter.
Fitness therefore acts as a proxy for adaptation quality, not just performance.
How Fitness is used in darwintIQ
darwintIQ evaluates each trading model over a recent rolling time window that moves forward with the market. Within this window, multiple behavioral aspects are measured — such as return stability, drawdown structure, and distribution consistency.
These components contribute to a model’s overall fitness score.
Models are then ranked by fitness inside the evolving population.
Because the window continuously shifts, fitness is always tied to the present market regime. Models whose behavior remains coherent under current conditions maintain high fitness and are more likely to persist across generations. Models whose performance depends on unstable or fading structure lose fitness and disappear.
In this way, darwintIQ uses fitness exactly as evolutionary systems do:
as the signal that determines which behaviors survive in a changing environment.