This is for planning and risk management. I am stuck on the following thoughts -

  1. Back-test the trading strategy for a period similar to the one you expect and then project.
  2. Do the above using empirical distribution of the back-test period, similar to FHS.
  3. For portfolio, GARCH estimate variance into the future time steps, run MC using empirical distribution. This is like to except time varying volatility, also, you are doing it at each asset level. May use copula and draw correlated noise.
  4. Assume regardless of the market condition that trader or manager will find suitable trades and use his Sharpe ratio or %gain. Here draw-down will be difficult to handle. May be assume he will stick within his risk limits.

Please provide references with your answers.


closed as unclear what you're asking by Ric, Quantopik, olaker Jul 6 '15 at 11:15

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.


Backtesting on a past realization does not provide any meaningful "estimate", as the variance of the "estimate" would be undefined.

More meaningful would be to make distributional assumptions and get estimates through extensive Monte Carlo simulations.

Clearly, the estimates that you get would be "meaningful" under your specific distributional assumption, and, in practice, under mild deviations from these assumptions.

(It remains understood that the mkt has no obligation whatsoever to "follow" your assumptions and distributions, and that is just a work model for you to get some indication. In addition, keep in mind that a simulated environment does not include the effect of your orders, which may and, with significant orders, actually is not negligible).

  • $\begingroup$ Please give me references for your quotes. 1. why backtest variance estimate will be infinite? How do you get distributional assumptions through extensive MC? $\endgroup$ – user12348 Jul 9 '14 at 14:08
  • $\begingroup$ Well, I am not using references apart myself. If you have an estimate based on 1 (unique) past realization, clearly is unusable because the variance can't even be estimated. Through extensive MC simulations you get estimates, not assumptions. The assumptions are what you "assume" when creating the "simulated" scenarios. Clearly, the mkt does not care and has no "obligation" whatsoever to follow your "models" (as and the mkt makers, on the other hand, do try to to take your money from you with all sort of tricks :-) $\endgroup$ – Pam Jul 11 '14 at 0:17

Not the answer you're looking for? Browse other questions tagged or ask your own question.