Numbers feel authoritative. When a metric displays a large value, the natural human response is to treat it as meaningful — a high count suggests something significant has happened many times, and many times sounds better than few times. This intuition is correct in many contexts. In trading system design, it's a trap that I walked straight into during the early weeks of running the secondary signal layer, and it took a specific real-world case to make the problem undeniable.
This post is about what happened when a stock generated a crossing count above 170 in a single session, what that actually meant, and what it changed about how I think about quantitative signal metrics in general.
The Day the Counter Hit 178
It was a Wednesday morning about three weeks after the secondary signal layer had gone live. I opened the dashboard at around 9:45 and saw a stock — a capacitor manufacturer — showing a secondary signal count of 178. The primary signal had fired just after 9:21. In the roughly twenty-five minutes since, the secondary signal had crossed the 120-period moving average 178 times.
My immediate reaction was that something was wrong with the code. 178 crossings in 25 minutes seemed impossible. I pulled up the stock's chart on my brokerage platform to see what was happening.
What the chart showed was immediately obvious and, in retrospect, exactly what I should have predicted: the stock's price was sitting almost perfectly on top of the moving average, oscillating back and forth across it with tiny price movements — a few hundred Korean won up, a few hundred won down, back and forth, again and again. Each microscopic upward tick that took the price fractionally above the average was being recorded as a crossing. Each subsequent downward tick that took it back below reset the state. The 178 crossings weren't 178 meaningful momentum shifts. They were 178 instances of a price twitching back and forth across a reference line it had no conviction to leave.
The trading value and trade intensity multiples at each crossing were consistent with this interpretation: both were running at or below 1.0x relative to baseline. The stock wasn't doing anything with the crossings. It was consolidating.
Why the Number Felt Meaningful When It Wasn't
The 178 reading triggered something real in my psychology even though I intellectually knew better. A three-digit number on a live dashboard has presence. It implies accumulation, repeated confirmation, a signal that keeps regenerating itself. The mind wants to interpret it as "this stock has shown bullish behavior 178 times today" rather than the more accurate "this stock has been stuck at its moving average for 25 minutes."
This is a specific manifestation of a broader problem in quantitative analysis: metrics that are designed to count occurrences of a meaningful event become misleading when the event they're counting can occur for meaningless reasons. The secondary signal crossing counter was designed to track meaningful momentum crossings. It was counting all crossings regardless of whether they were momentum-driven or consolidation noise.
The number itself was not wrong. The interpretation was wrong, and the design of the display was making that wrong interpretation natural rather than difficult.
Three Other Cases That Reinforced the Pattern
Over the following two weeks, I started paying closer attention to the distribution of crossing counts across all stocks that generated secondary signals. The pattern that emerged was consistent enough to constitute a reliable heuristic.
The first case was a pharmaceutical stock with a crossing count of 94 by noon. The trading value multiple at each crossing was averaging around 0.3x — meaning each crossing was happening on about 30% of the stock's normal post-signal trading volume. The stock spent the entire afternoon trading in a narrow range and closed essentially flat from where it had been when the primary signal fired. High crossing count, low volume, no direction.
The second case was the opposite in almost every way: a semiconductor equipment stock with a crossing count of just 4 by the time I was reviewing the dashboard at 10:45. But each of those 4 crossings had occurred with trading value multiples above 2.5x, and the price had advanced meaningfully after each one. The stock went on to close up nearly 12% from the primary signal level. Low crossing count, high volume at each crossing, strong subsequent move.
The third case was instructive for a different reason: a logistics stock that showed a crossing count of 31, which I initially treated as moderate — neither a clear consolidation signal nor a particularly strong confirmation. When I looked at the timing of the crossings, they were clustered: 28 of the 31 crossings had occurred in a single twelve-minute window early in the session when the stock was clearly oscillating. The remaining 3 crossings had occurred spread across the next two hours, each with meaningfully higher trading value. The clustered 28 were noise. The spread-out 3 were signal. The aggregate count told me almost nothing useful; the timing distribution told me quite a lot.
What the Right Questions Actually Are
After working through these cases, I shifted how I use the crossing count in my manual review process. The count itself is no longer the primary question. The questions I now ask when reviewing a secondary signal are:
First, what is the average trading value multiple across the crossings? If it's below 1.0x, the crossings are probably consolidation noise regardless of how many there are.
Second, how are the crossings distributed across time? A count of 50 crossings spread across a single fifteen-minute window is almost certainly sideways noise. A count of 8 crossings spread across two hours, each with elevated volume, is a different situation entirely.
Third, is the crossing count accelerating or decelerating? A count that grew quickly in the first hour and then slowed down often indicates the consolidation phase is ending and directional movement is beginning. A count that keeps growing at the same rate all session suggests the stock is still going nowhere.
None of these questions can be answered by looking at the crossing count alone. They require looking at the timing of crossings and the volume at each one — which is why the dashboard was updated to display the time of the most recent crossing and the trading value multiple, rather than just the count.
What This Revealed About Dashboard Design
The 178 case ultimately changed how I think about dashboard design more broadly. A number on a display doesn't just convey information — it shapes the interpretation of that information. Displaying a crossing count of 178 prominently creates an implicit claim: that 178 is a meaningful quantity. The design was making a misleading claim without stating it explicitly.
The fix was partly about adding context alongside the number — the trading value multiple and timestamp make the number harder to misinterpret. But the deeper lesson is about what the dashboard is for. A good trading dashboard isn't a scoreboard. It's a tool for forming accurate beliefs about what the market is doing right now. Every metric on it should be making that task easier, not harder. When a metric consistently leads toward inaccurate beliefs — even if the number itself is technically correct — the metric's design needs to change.
Distinguishing Signal From Noise Is the Core Problem
Looking back at this episode and the broader pattern it exposed, the fundamental challenge is one that every signal detection system eventually has to confront: distinguishing events that are meaningful from events that merely look like they satisfy the definition of meaningful.
A moving average crossing, in isolation, means the price moved from below to above a computed line. That's a mechanical fact. Whether it means anything depends entirely on the context — the volume at the crossing, the history leading up to it, the broader market environment, the behavior of the crossing count over time. None of that context is captured in the count itself.
This is why building a signal detection system isn't primarily a technical exercise. The technical parts — the code, the database, the dashboard — are straightforward once you know what you're trying to build. The hard part is figuring out what's actually meaningful and what's noise, which requires watching real market data over time, forming hypotheses, testing them against more data, and being willing to update your interpretation when the data contradicts your initial assumptions.
The 178 crossing count was a good reminder of that process. It looked like signal. It was noise. The difference was visible only when I stopped looking at the number and started looking at what was actually happening in the market.
Today's Investing Insight — The Multiple Comparisons Problem in Trading Research
One of the more subtle statistical traps in trading system development is the multiple comparisons problem. When you test many different indicators, parameters, and combinations against historical data, some of them will appear to work purely by chance — the same way that if you flip a coin enough times, you'll eventually get ten heads in a row. The more tests you run, the more likely you are to find something that looks meaningful in the data you tested it on but fails in new data. This is related to overfitting, but it's slightly different: the issue isn't that your model is too complex — it's that you ran enough tests that one was bound to look good. The safest guard against this is to form a hypothesis based on a real market observation before testing it, rather than searching through combinations of indicators until something appears to work. The secondary signal crossing count redesign described here started from a real observation — consolidating stocks produce many low-quality crossings — rather than from data mining. That grounding in an actual market mechanism makes it more likely to hold up over time.
---
This post documents a personal journey of building an algorithmic trading system and is not a recommendation of any specific stock or strategy. All signal metrics described here are summaries of historical market data and do not predict future price movements. All investment decisions and their outcomes are the sole responsibility of the investor.
Comments
Post a Comment