Systematic Does Not Mean Automated
· #systematic-trading #trading-psychology #research-method #execution-parity #volume-profile
The first time someone asked me what systematic trading meant, I gave the lazy answer. I said it meant the computer places the orders.
That answer is wrong in a way that took me a while to see properly. Automation is the cheapest part of the whole thing. Anyone can write a script that clicks buy. The hard part, and the part that actually decides whether you can trust the thing, is everything that has to be true before the script is allowed to click anything.
What does systematic actually mean, if not automated?
The working definition I ended up with has four parts, and none of them mention a computer.
The logic is explicit. The rules are testable. The assumptions are visible. The result can be reproduced.
That last one is the quiet killer. If I rerun the same logic on the same data and get a different set of decisions, I do not have a system. I have something fast that I do not understand. So reproducibility is not a nice property I add at the end. It is the gate.
No magic, no mystery, no hand waving. That is the standard I keep aiming for, and I miss it more often than I hit it.
What was the hard part, if it was not learning another indicator?
It was watching how easily my own mind distorted a judgement I had already made.
Hope turns into bias. Fear turns into hesitation. Frustration turns into overtrading. None of those feel like errors while they are happening. They feel like reading the market.
The conclusion I reached was not "trade without emotions." That is not available to anyone. It was "no emotional execution." Emotions can stay. They just do not get to place the order.
That is the specific job I gave the software. In Master Volume Profile, my own volume profile approach for short duration setups, the coded caps and the ATR based volatility veto exist to cover exactly the moments where a human would hope, hesitate, or push one more trade. Not to remove human judgement, just to stop bad judgement from executing.
Why do most indicators fail to earn their place?
EMA, RSI, ADX and the rest are useful. They are also not magic. They describe state, and describing state is not the same as creating edge.
So I stopped treating them as a belief system and started treating them as candidates. Every feature has to answer what it measures, when it applies, and what would get it dropped. In the Dquants engine I built, a feature that cannot demonstrate a contribution gets cut, no matter how familiar it looks on a chart.
The related question is timing. A lagging signal can still be useful, but if the only thing it tells me is that a move already happened, it is too late to be entry logic. That is not a small distinction. A lot of comfortable looking systems are built on confirmation that arrives after the opportunity.
What replaced candle reading for me?
Structure. Where value has been accepted, where price is being rejected, what the auction context around a move looks like, how much runway exists before the next area of interest.
That framing explains why a move continues or fails far better than the shape of the last three candles does. It is also why Master Volume Profile reads volume profile structure rather than candle patterns. Value areas and volume nodes are on the chart before the move, not after it.
I want to be careful here, because this is where people usually overclaim. Structure makes a hypothesis worth testing. It does not exempt it from testing. Plenty of structural ideas I liked did not survive.
Why is parity the part almost nobody talks about?
This is the piece I underestimated the most, and it is the one I would now put near the top.
Research and execution have to tell the same story. The C++ research engine inside Dquants and the MQL5 side of Master Volume Profile have to produce the same result on the same data. If they do not, I genuinely do not know which one I am supposed to believe, and every validation I ran upstream stops protecting anything.
Small mismatches are easy to wave away in a notebook. They stop being small when money is on the line. Parity is not a side task on the way to deployment. Until it holds, a research result is still a draft.
Why do costs decide more than the entry rule?
Because costs are a filter for truth.
Plenty of ideas look strong before costs and fall apart after them. On short duration work especially, spread and slippage are not a rounding error sitting next to the signal. They are part of the signal. So the habit I built was to stress the edge before celebrating it: push spread and slippage past realistic levels in a cost sensitivity sweep and watch where the thing breaks.
Finding the breaking point early is much cheaper than discovering it with real money later.
How do I stop myself from believing an early result?
Mostly by refusing to let an early signal graduate on its own.
An early signal is a reason to keep investigating. It is not a reason to lock the idea. Before I let myself believe more, the parity check has to hold between backtest and demo execution, and the metrics have to hold up as metrics: Sharpe, Calmar, expectancy, drawdown, with deflated and probabilistic Sharpe as a guard against reading luck as skill.
Data is a useful mirror precisely because it does not care what I hoped would happen. Sometimes that is encouraging. Often it is humbling. Vibes do not compound.
This is also why I try not to overpromise. When I say something is promising, I mean it still needs validation. When I say something is strong, I mean it earned that word in a test. Master Volume Profile has no verified live track record yet, only design choices and tests I can show you, and saying that plainly protects the reader as much as it protects me.
What is the honest takeaway?
Real progress in this work is thin and technical. Better validation, cleaner exports, stronger parity, more honest reporting. None of it is exciting to describe, and all of it compounds.
If you are an engineer looking at markets, my honest expectation is that the automation will take you a weekend and the rest will take you years. Systematic is not the bot. Systematic is the discipline of being able to explain, test, and reproduce every decision the bot makes, and being willing to throw the whole idea out when you cannot.
This article was composed from the KenKem build-log series. Educational purpose only. Not financial advice. Past performance does not guarantee future results.