pedma

work for this week is to reduce turnover.
there's 2 levers I can try to smooth.
a) the actual signal
b) the volatility forecast
I am starting off with the signal because as we can see it's pretty bouncy even on a portfolio level. I've done some work in the past to smooth it over but truth to be told I haven't looked at this with the detail I should've.
signal still pretty bouncy and I am sure I can make it a bit better without giving too much of what the signal provides.
the volatility estimate is an option too, but its something I haven't thought carefully about the drawbacks of making it smo

another edge case bug i caught today.
the other day I was moving funds around, and apparently didn't left enough margin on an extended sub account.
as it was executing the positions, it ran out of margin, and partially filled positions were recorded as "complete" due to how that information is received from the exchange API.
i have a method to handle partially filled positions and they get appropriately categorized as such and I get warnings BUT when it comes to partials due to margin controls, it apparently breaks.

watching odyssey today at mid night in IMAX. pretty bad time but was the only spot available in IMAX this week that wasnt bottom row looking up to the screen.
considering I go to bed at 9-10pm each day this should be fun. wouldnt be the 1st time I sleep through a movie.

about 42 recurring issues that my agent was reporting each day. these generate 1000's of lines.
going down the list of highest occurrence to lowest, and already dropped it from almost 4k warnings a day to 11.
still on #15 but will clean them all up, just noise on the backend.

i think to be able to handle well the swings of full-time trading people need to look at the ratio of avg monthly return vs cost of living.
mine right now sits between 5-10x ratio. meaning for each month, on avg, I've been able to produce 5-10 months of expenses.
obviously these are figures from the past, and it's averages, some months are red, etc, and the idea that I'll maintain edge is always speculative.
there's periods where that 6 month average went negative but i think with the changes i've made since that period, that those periods should be shorter, but who knows.
keeping this gap bet

my portfolio tilt is turning a bit green the past few days.
what's interesting is the disparity between the net notional sizing which is -82% short and the net vol adjusted sizing which is 7.8% long.
meaning that although my portfolio is tilted slightly net long, on a vol adjusted basis, the short side asset's vol is much lower than the long side vol, taking much larger notional weight.
what could hurt me is if this thing were to flip and a lot of this trash that im short began getting attention, and everything started pumping together, i'd be in for a pretty nasty few days, and the longs I cu

as regulatory environment tightens in Europe I am being more careful by saving my data more regularly, reconciling my trades, making sure that everything is explainable in a future audit.
for example, on hyperliquid i can only go back about 10k trade records. for me that lasts about 3 months and im already not able to go back any further. if I had to prove transactions from earlier periods and I did not have a database with the entire history, I would not be able to prove the operations I am reporting.
I don't have to go back as I have every single trade recorded in a database, but this is the

a perma bid in equities and we choose to trade this.

now some turnover stats. I think I am turning over a bit much (283x annualized) even though I did some work to get this down in the past.
i used the highest cost brackets on each exchange, although not all is taker volume. slippage cost is real from live stats.
this is something I have to look into again. I think i can do better in my signals without giving up too much edge.

been in such a great winning period lately that it seems I forgot what's like to be in a drawdown. now that I am in one again, here we are back at building dashboards to monitor whats going on.
here's the port performance and IC across all the features on two of my portfolios. definitely not the best looking period for the features I am running now.
the first one is pure momentum. you can see how correlated these all are.
I've been working on another (second pic) that will include different features and you can see that some are already doing better than others and there's more dispersion acr

when i am working on new features, i sometimes come up with different variations of the signal I am working on. but testing them in bulk through entire datasets takes some time, so i just built this queue pool where i can be working on new features, while the others run in the background, and then i get a warning when they're done.
just a neat way of not getting stuck waiting for results. also i'll get alerts when its completed so if i have a large set i can just go off desk and chill out while I wait for the results.
this then follows a pipeline where if they have acceptable results based on