# Risk Management > Risk management is about changing the payoff function g(.) rather than making "good forecasts". > - Statistical Consequences of Fat Tails ### Ideas * Probabilistic sharpe ratio * Strategy risk vs portfolio risk * Dependencies - what is our strategy dependent on? E.g. it could be that when we get good forecasts we do *very* well, but when we get bad forecasts we get crushed. This is a dependency on forecasts. * Example: Our strategy is heavily based on a solar forecast. If that forecast is *wrong* then we get crushed. * Perimeter of our training dataset. E.g. what scenarios make up the boundary of what we have seen before? And then what is outside of that? It could be that we just don't have much data regarding temperatures above 105F and if that is reached we are not as confident in our system. * For any upcoming day it should be compared (along certain dimensions) to past that we have backtested against. If we are way outside of those days we should at least be alerted. * How does the system perform near "the edge" of our training data? * What scenarios would lead to the worst performance? * How much data could have been leaked via running many backtests? * How tied to past dynamics is it? If the next month is different than the past due to, say, batteries being added to the grid, what does that mean? A strategy is made up of: 1. A model (that generates predictions) 2. A scorer that takes these predictions and effectively ranks them 3. A portfolio builder that creates bids based on our scores and other constraints A strategy exists within the space of all strategies, $S$. A **robust** strategy is one that, if perturbed slightly, would still perform well. An example of a perturbance would be a change in the underlying data: e.g. the addition of noise. --- Date: 20230808 Links to: Tags: References: * []()