macrocephalopod

and subject the overall portfolio to a large series of extreme stress tests based on large directional moves or basis moves, ensuring that it stays within acceptable tolerances under a wide range of assumptions about how markets behave

rather than building some super complicated highly overparameterized stochastic volatility model and then "believing" this model in some deep sense reflects reality

typical overall end point might be, for example: in a market environment reflecting an average overall reward/risk ratio, having a forecast 10% annualized PNL volatility and 2% daily 99% VaR in the book increasing overall risk proportionally to the overall risk/reward ratio

allocating risk across strategies roughly proportionally to adjusted/handicapped reward/risk ratios, but with significant downsizing of strategies that are expected to have material positive correlations with other strategies in the book

and material upsizing of strategies that are convincingly complementary to the rest of the book and/or are very likely to make money if the shit hits the muthafuckin fan

how you manage this day to day depends on how liquid the products you trade are and what kind of transaction costs you incur trading them. in pretty liquid stuff, you are continually managing towards what you see as the optimal portfolio (as discussed above), never too far away

in more illiquid markets (long term rates etc) you might be thinking about every day what are the handful of trades you can do that best transform your risk profile in the direction of where you want to be

mechanical stop losses generally are inappropriate here - in a dislocations oriented strategy, pnl tends to be mean reverting, and expected returns highest after a drawdown, as long as risk is well controlled (not short convexity)