What medium-frequency trading means for us
26 October 2026 · Annova Capital
Medium-frequency sits between two better-known styles. High-frequency firms compete on microseconds and hold positions for moments. Long-horizon funds compete on judgment and hold for months. Medium-frequency strategies hold positions from minutes to days, so the edge comes mostly from the quality of the signal and the care taken in execution, not from raw speed.
Speed still matters. A slow system leaks money through poor fills, and an unreliable one leaks it through missed orders. That is why we build and run our own low-latency infrastructure. The aim is to be fast enough that execution never gets in the way of the idea, not to win a race measured in nanoseconds.
It also changes how we do research. With holding periods of minutes to days, there is enough data to test an idea properly, and enough time for costs to matter. A signal that looks good before costs and disappears after them is not a signal.
Backtests are hypotheses, not results
26 October 2026 · Annova Capital
A backtest answers one narrow question: would this rule have worked on this history? It does not tell you the rule will work next month. Every extra parameter, and every rerun with a small tweak, gives the data another chance to flatter the idea. After enough tries, something always looks good.
We treat a backtest as a hypothesis and try to break it. Does it survive data it was never tuned on? Does it survive realistic costs, slippage and the delay between signal and fill? Does it still work if we nudge each parameter slightly? If a result depends on one exact setting, we assume it is noise.
Only after that does a strategy go live, and then at small size with tight limits. Early live results are compared with what the research predicted. When they disagree, we assume the research is wrong until shown otherwise.
Risk is an engineering problem
26 October 2026 · Annova Capital
Most trading losses that make the news come from systems doing exactly what they were told, at a scale or in a situation nobody planned for. Good intentions do not stop that. Controls built into the software do.
In our platform, every order passes through checks before it reaches a broker: size limits, price sanity checks, exposure and margin limits, and a cap on how many orders can go out in a given period. Limits are set per strategy and for the firm as a whole. Monitoring runs continuously, and a kill switch can halt a strategy or the whole system with one action.
We also assume parts of the system will fail: a broker connection drops, a data feed stalls, a process crashes. The design question is not how to prevent every failure. It is what the system does in the first ten seconds after one.
