Why the Search Space Explodes
A conventional deterministic optimizer can lock the economic assumptions in place and find the best set of balancing decisions for that one scenario. A Monte Carlo simulation can do the opposite: lock the balancing policy in place and run, say, 10,000 trials with different random assumptions to show a distribution of possible outcomes.
Both are powerful shortcuts, but neither searches the complete joint problem. If there were 100 uncertain macroeconomic factors and each could take only 20 values, the scenario space would contain 20^100 combinations — far more than could ever be enumerated. Adding the possible portfolio decisions multiplies the problem again.
The state of the art was therefore commonly to do one of two things: run Monte Carlo while holding the balancing decisions fixed, or run a balancing optimizer while holding the random factors fixed. The first measured risk around a predetermined policy. The second optimized for a predetermined world. What the client wanted was to improve the policy because the world was uncertain.
Optimizing Under Uncertainty
We represented the portfolio’s assets, liabilities, hedging instruments, transaction costs, and operating constraints together with stochastic influences such as rates, prices, demand, and other macroeconomic variables. A candidate strategy was evaluated across a sample of plausible futures rather than against one point forecast.
That changed the objective. The model was not simply maximizing the expected outcome. It could weigh downside exposure, tail results, liquidity, hedging cost, and the stability of a strategy across scenarios. A slightly lower average return could be preferable if it substantially reduced an unacceptable loss in adverse conditions.
Decisions That Survive Uncertain Futures
The system gave decision-makers a policy with an explicit risk profile, not a single answer disguised as certainty. They could inspect how the recommended balance behaved across thousands of trials, identify which assumptions mattered most, and understand the cost of protection against adverse outcomes.
Monte Carlo remained a shortcut (10,000 trials are microscopic beside 20^100 theoretical combinations), but a well-designed sample can reveal enough of the landscape to make better decisions. The point was not to solve every possible future. It was to search for a more resilient strategy without pretending that the future had been locked in place.
One Technology developed a method that could optimize in the presence of stochastic influences. Instead of treating simulation and optimization as separate exercises, the search used simulated futures to compare and improve balancing decisions. That integration was the distinctive technical contribution: uncertainty influenced which decisions the optimizer selected, rather than being applied only after the decisions had been fixed.
See also: Milk
