One of the most valuable — and humbling — experiences in building this system was discovering that my initial interpretation of the signal data was not just slightly off, but directionally wrong. I had spent weeks watching charts, comparing signal timestamps to subsequent price action, and building a mental model of how the system was performing. That mental model turned out to be built on a systematic perceptual error. This post is about that error, how I found it, and what it actually revealed about where the signals were firing.
What I Thought I Was Seeing
About six weeks into running the live signal system, I started reviewing historical charts from the days when confirmed signals had fired. My process was straightforward: pull up the chart for each stock, find the approximate time when the signal had been generated, and assess whether the entry would have been good or bad based on what happened next.
My initial conclusion after this review was troubling. Looking at chart after chart, the signals appeared to be firing near the top of moves. I could see on multiple charts that the price had been in a strong uptrend, and the signal appeared to coincide with — or just after — the highest point of the day. It looked like the system was consistently arriving late, tagging the peak and generating a signal precisely when the move was exhausted.
I was ready to conclude that the signal detection logic had a fundamental flaw. The system was identifying breakouts after they had already run their course.
The Mistake in How I Was Reading the Charts
Before making any changes to the system, I decided to verify my impression more carefully. I went back to the trade logs and extracted the exact timestamps — to the second — of every confirmed signal from the previous two weeks. I then annotated each chart with a vertical line at the precise moment each signal fired.
What I saw when I did this properly was almost the opposite of what I thought I had seen.
The vertical lines were landing near the lows of the day — or at most, in the middle third of the day's range — not at the peaks. In chart after chart, the signal had fired during or shortly after the pullback, just as the price was beginning to recover. The subsequent price action often continued higher for another 10%, 20%, or more after the signal.
What I had been doing wrong was reading the charts without the timestamp reference and allowing my eye to be drawn to the most visually prominent features — the peak and the surrounding area. Looking at a completed day's chart, the eye naturally lands on the highest point. When I knew approximately when the signal had fired, my memory and perception had been anchoring to the peak, not to the actual signal point. A signal that fired at 10:15 on a stock that peaked at 11:30 looked, in retrospect, like it had fired "near the top" — even though from the perspective of 10:15, the move hadn't happened yet.
This is a textbook example of hindsight bias applied to chart reading. The chart shows you the complete story, including the ending. Your brain fills in the narrative backward. A signal that fired before a large move gets mentally relocated toward the end of that move because that's where the story is most dramatic on the chart.
What the Corrected Analysis Actually Showed
Once I was reading the charts with precise timestamps, the picture changed substantially. Across the general cases I reviewed, signals that fired when the precise timestamp was near the day's low — meaning the signal came during or just after the pullback — were followed by meaningful continuation moves a majority of the time. Signals where the timestamp showed the system had fired near an established high or after a large move had already occurred were more likely to reverse.
This was actually confirmation that the wave-pullback-breakout structure was doing what it was supposed to do. The system was designed to identify pullbacks and catch the re-breakout — and the data showed that it was, in fact, doing that. The apparent problem I had identified from my initial casual review wasn't real.
The lesson wasn't about the system. It was about how I was analyzing it.
How I Changed My Review Process
After this experience, I established a strict protocol for reviewing signal outcomes: every analysis must start with the precise timestamp from the trade log before looking at the chart. The timestamp gets marked on the chart first. Only then do I assess what happened before and after.
This seems like an obvious thing to do, but it isn't how most people naturally review charts. The intuitive approach is to look at the chart first and find where the signal "was." The problem is that the chart, looked at as a whole, shapes your perception of where the signal was in ways that are systematically biased toward the most dramatic price action.
Precise timestamps anchor the analysis to what was actually known at the moment the signal fired — which is the only perspective that matters for evaluating whether the signal logic is working correctly.
Today's Investing Insight — Hindsight Bias in Trading
Hindsight bias — the tendency to believe, after the fact, that an outcome was more predictable than it actually was — is one of the most consistently documented cognitive biases in behavioral economics research. In trading, it shows up constantly. After a stock makes a large move, experienced traders and beginners alike look back at the chart and find the "obvious" signals that predicted the move. What they're actually doing is pattern-matching backward, selecting the signals that happened to precede the move and ignoring the equal number of identical signals that preceded nothing. This is why backtesting on charts you've already seen is an unreliable way to validate a strategy. The patterns that look predictive in hindsight are often no more predictive than the patterns that led nowhere — they're just more memorable because something happened after them.
---
This post documents a personal journey of building an algorithmic trading system and is not a recommendation of any specific stock or strategy. Past signal outcomes do not predict future results, and all investment decisions and their outcomes are the sole responsibility of the investor.
Comments
Post a Comment