Closed druce closed 3 years ago
Aside: I've also had trouble pip installing prophet although upgrading pip and installing pandas first seems to help.
very cool stuff!
this is my mad science project, so trying a few more optimizers was very helpful https://github.com/druce/swr/blob/master/optimize.ipynb
basically the idea is optimize safe withdrawal … write a function that takes a stock/bond allocation, and a couple of withdrawal rates, evaluates how it would do historically (or Monte Carlo) using CRRA certainty equivalent cash flows (real cash flows discounted for volatility). gives reasonable answers, historically works a little better than Bengen '4% rule'. i.e. can spend more without running out if you tolerate a little variability. for simple rules gradient-based Powell or L-BFGS-B works too. Anyway will blog about it sometime!
will follow your stuff with interest!
On Mon, Feb 8, 2021 at 9:06 AM Peter Cotton notifications@github.com wrote:
Merged #6 https://github.com/microprediction/timemachines/pull/6 into main.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/microprediction/timemachines/pull/6#event-4303604983, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJDX3VHF34Y3XEPYI4ZMI3S57VWPANCNFSM4XGTXAZA .
Very cool. If you want we could include it as one of the objectives - always looking for real-world objective functions rather than the fake ones.
thanks Peter! still very much a work in progress but eventually will blog about and maybe can provide a canonical case/benchmark!
On Tue, Feb 9, 2021 at 9:52 AM Peter Cotton notifications@github.com wrote:
Very cool. If you want we could include it as one of the objectives - always looking for real-world objective functions rather than the fake ones.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/microprediction/timemachines/pull/6#issuecomment-775997286, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJDX3UEWIRGCSPGGTPLSSDS6FD33ANCNFSM4XGTXAZA .
add requirements.txt for optimizers
includes optimizers that are imported but not currently used by alloptimizers.py, e.g. fbprophet, hyperopt
locally I can install fbprophet with conda install fbprophet but not pip install fbprophet . but probably a local issue and in any event maybe not used