Underfitting is the opposite failure to overfitting and gets far less attention in trading, partly because it is less embarrassing: a flat backtest is quietly abandoned rather than traded and blown up.
It shows up as poor results both in sample and out of sample, with residuals that still contain obvious patterns. A single global volatility threshold applied across equities, bonds, and crypto will underfit, because the sensible level differs by an order of magnitude per market.
In practice, the bias-variance-tradeoff means you move between the two failures. Trading sits far enough toward the noisy end of the spectrum that erring toward the simple model is usually right, but not always.
Related: overfitting, bias-variance-tradeoff, regularisation