The idea is to reduce risk when the curve turns down and increase it when it recovers, as though the equity curve were a price series with trends. Whether it works depends entirely on whether your results are actually autocorrelated, which for most discretionary traders they are not.
Where it helps, the mechanism is usually behavioural rather than statistical: cutting size during a bad stretch keeps the trader executing and prevents the escalation that would otherwise arrive. That is a real benefit, but it should be named honestly.
If you do it, make it a rule with numbers - thresholds, size levels, and the conditions for returning to normal - and test it against simply keeping size constant. Done by feel, it becomes undertrading after losses and position-size-creep after wins.
Related: drawdown-psychology, position-size-creep, undertrading, risk-per-trade