Create function/method which combines investment-funnel modules to future-test for different life cycle and risk appetite/preference about investment-funnel

[mikkelbechmogensen]

The 5 steps has been done. Right now, however, the future testing is done on 1000 scenarios, which is done in a cvxpy enviroment, that seeks to minimize costs of rebalancing. The issue with this approach is that it takes approximately 10 minutes to run the future analysis test for each risk level, which is very time consuming, and doesn't work well, if it has to be implemented in the investment funnel page for dynamic use.

[tschm]

I don't understand why you would need 10 years of data to estimate a covariance matrix. This should be done in a moving manner, e.g. as discussed in the paper by Johanson, Boyd et al. Please don't estimate a matrix over an interval and then use it to run portfolio optimization over the same interval. Your results will look amazing but you have a forward looking bias

[VanekPetr]

Hi @tschm, thank you for your comment. You are right, that estimating a matrix over an interval and then use it to run portfolio optimization over the same interval would be bad idea. The idea here is different, I just described it poorly.

The idea is to estimate covariance matrix based on 10 years of data (but should be configurable over frontend) and then just test sensitivity of the (optimal, for different levels of risk targets) portfolios to different possible future returns.

The second step (more advanced and not in the scope yet) then would be to estimate the covariance matrix on the test dataset and backtest it for different (out of the train sample) dataset. For "lifecycle" backtest, where we would like to change risk levels over time periods, we would wish to have much more data and therefore also task "Solve our survivorship bias" is in the backlog.

Let's discuss it @mikkelbechmogensen.