Setting the Trading System's Goal - Absolute Return ~10%/yr, Scrapping the Always-Fully-Invested Rule
Yesterday I wrote that "stay positive even when the index drops" and "never leave cash idle" couldn't both be true, and that choosing between them was my call. Today the owner decided: this is a profit system aiming for a long-run absolute return of around 10% a year net of costs, and the rule that it must always be fully invested is gone. As a first step, an intraday filter that briefly defers trades on sharp spikes and drops went live on the real account.
This is the English version of a post originally written in Korean for my algorithmic trading system devlog(new tab).
Yesterday's open question got an answer
Yesterday's post(new tab) ended on an uncomfortable question.
After measuring the performance metrics honestly again, the stock-selection edge turned out to be essentially zero, and almost all of the loss came from the market falling as a whole.
So "stay positive even when the index drops" and the design rule "never leave cash idle, always stay fully invested" simply couldn't both be true — and choosing between them was my decision, not something code could resolve.
Today I made that call.
This is a "profit system," and the yardstick is absolute return
First I nailed down its character. This is not a research-and-learning toy; it's a system meant to make money.
The success criterion is absolute return, not return versus the index. If, net of all trading costs including electricity, it earns around 10% a year over the long run, I count that as success.
It doesn't have to hit exactly that number every year. With compounding, a long-run average near that figure is fine, and a somewhat lower year or a somewhat higher year is acceptable.
Electricity goes into costs at its actual rate, but hardware purchase price and depreciation don't — "it's aging gear anyway, as long as it makes money."
Changing the yardstick changes the first question I ask of performance. It used to be "did it beat the index?"; now it's "is it reliably positive after costs?" Index-relative numbers are only a secondary read.
Scrapping the "never leave cash idle" rule
The paired decision was retiring an old rule: "don't leave idle cash."
The original logic was that the money you put in the account is itself a decision to deploy it, so holding cash violates that intent. The system was therefore designed to stay close to fully invested.
The problem is that this structure has no way to be positive in a falling market. Long-only and always fully invested means that when the market drops, there's nothing to do but drop with it.
So I decided to allow intentional cash holdings. Cutting exposure and sitting in cash when the market is bad is now a legitimate option, not a rule violation.
I'm keeping one distinction, though. Cash the signal deliberately leaves ("hold less for now") is fine, but cash that piles up unintentionally from a misconfiguration is still a defect. Adding new exposure logic to the real account also still has to clear the existing safety-freeze process.
First step — an intraday filter that briefly defers trades on spikes and drops
With the goal changed, the lever to pull right away was cutting needless trading friction.
So I added an intraday trade filter(new tab) to the real account.
The idea is simple. If a stock I meant to buy spikes sharply during the session, the buy is briefly deferred to wait for a pullback.
Conversely, if a stock I meant to sell drops sharply, the sell is briefly deferred to leave room for a bounce. The point is to cut the cost that leaks out through chasing rallies and panic selling.
The first version had several unnerving bugs. It double-counted sell credit, let deferred orders net against each other, and could open a reversed order when a deferral was cancelled.
Only after an urgent review — blocking the double-counting, the netting, and the reversed orders, and adding a separate intraday buy gate — did it go live. This is the layer where money actually leaves the account, so I scrutinize it harder than usual.
Along the way I also found and fixed a cash-buffer setting that had been partly dropped when it was carried over to the real account.
A few side cleanups
While resetting the goal, I also cleared up a couple of experiments that had been running in limbo.
The shadow experiment running an extra large model during the session got shut down. The conclusion was that it didn't improve the signal enough to justify its cost (intraday GPU load), so I ended it and kept only the conditions for restarting it as a guard.
I also switched the source of the price candles used for technical analysis from a paper-account feed to official exchange data. While at it, I retuned a threshold to cut false alarms that had been firing near limit-up and limit-down prices.
And from a code delta-review I'd handed to an external AI, I picked the high-importance findings and applied them in one batch. The resident processes affected were restarted together to pick up the changes.
Why it matters
Today's core wasn't a line of code; it was settling what this system is for.
Until yesterday the goal itself was self-contradictory, so no matter which signal I fixed, the foundation stayed shaky — "beat the index" and "stay fully invested" were fighting each other.
By resetting the goal to absolute return and the design to "cash is allowed," the direction of improvement finally converged. The question is no longer "how do I beat the index by more?" but "how do I cut costs and lose less in a down market?"
The trade filter is the first spade in that direction. Resetting the goal the day after fixing the measurement wasn't a coincidence — without measuring honestly, I'd have had no basis to make this decision at all.