The wave-pullback-breakout pattern was developed and initially tested during a period when the Korean stock market was trending upward with moderate volatility. The system performed well in those conditions. What I didn't know at the time was how much of that performance was attributable to the system's design and how much was attributable to the environment — a distinction that only became visible when the environment changed.
Over fourteen months of live operation, the KOSPI moved through three meaningfully different regimes: a trending upward phase, a highly volatile sideways phase, and a sustained declining phase. The same system, with the same rules, produced substantially different results in each. This post documents what those differences looked like, why they occurred, and what I changed in response.
Defining the Three Regimes
I'm using the word "regime" deliberately rather than "market condition" because I want to describe something more structured than daily fluctuation. A regime, as I'm using it here, is a sustained period — at minimum four to six weeks — where the market's dominant behavioral characteristics are stable enough to produce consistent patterns in the signal data.
The trending upward regime ran for roughly four months in the middle of the period I've been running the system. The KOSPI advanced approximately 12% over this window. Daily volatility, measured by the average ATR of tracked stocks, was moderate and stable. Institutional and foreign buying flows were net positive for the majority of sessions.
The volatile sideways regime lasted about three months. The KOSPI's net movement over this period was near zero, but the path was extremely jagged — individual sessions of plus or minus 2-3% were common. Average stock ATR was roughly double what it had been during the trending phase. The same stocks that had shown clean breakout patterns in prior months were now oscillating wildly within wide ranges.
The declining regime covered approximately two months. The KOSPI declined around 8% over this period, with a handful of sharp down days interspersed with partial recoveries. Foreign selling was the dominant institutional flow. Individual stock patterns were consistently interrupted by broad market pressure.
Signal Quality in the Trending Upward Regime
This was the environment the system was designed for, and the numbers showed it clearly. During the trending phase, confirmed signals produced continuation moves more than 65% of the time. A-grade signals were particularly reliable — I tracked only two A-grade signals in this period that resulted in immediate reversals rather than continuation. The average continuation move size on A-grade signals was meaningfully larger than in subsequent periods, frequently exceeding 5% within the two-hour measurement window.
The mechanism behind this performance is straightforward in retrospect. When institutional and foreign money is flowing into the market consistently, stocks that generate the wave-pullback-breakout pattern tend to have genuine demand behind their breakouts. The buying that drives the initial move doesn't disappear during the pullback — it pauses temporarily, then resumes. The system was detecting exactly what it was designed to detect.
One specific observation from this period that I hadn't anticipated: signal quality was consistently better in the 09:15-10:00 window than in the 10:00-10:30 window, even during the favorable trending phase. The very earliest signals of the session had the highest continuation rates. My best hypothesis is that in trending conditions, the stocks with the strongest demand tend to show the pattern earliest in the session, before the general market noise increases. Later in the window, the stocks triggering the pattern increasingly included those responding to intraday momentum rather than genuine overnight demand.
Signal Quality in the Volatile Sideways Regime
The volatile sideways regime was the most instructive period for understanding the system's limitations, because it was the environment where the rules continued to generate signals while the underlying conditions that made those signals meaningful had deteriorated most severely.
The continuation rate on confirmed signals dropped to approximately 42% — below 50%, meaning the majority of confirmed signals were followed by reversals rather than continuation. A-grade signals were no longer reliably differentiated from B and C grade signals in terms of outcome — all three grades produced continuation rates within a few percentage points of each other.
The problem was structural. High volatility meant that the ATR-based pullback thresholds had expanded significantly. Stocks were pulling back by larger percentages before qualifying for stage two confirmation, which meant the pullback phase was taking longer and running deeper. By the time a breakout was confirmed, more time had elapsed since the initial wave formation, and the demand that had driven the initial move had partially dissipated.
High volatility also meant that the breakout itself was less meaningful as a signal. In a low-volatility environment, a stock crossing its reference high by 1% with strong volume is a notable event. In a high-volatility environment where stocks routinely move 5-8% in a session, a 1% breakout is barely distinguishable from normal oscillation.
The single most useful adaptation during this period was reducing position size by approximately 40% compared to the trending phase and significantly raising my internal bar for acting on a signal. During the trending phase, I would act on most B-grade signals that cleared my manual review checklist. During the volatile sideways phase, I effectively required A-grade signals with strong participant confirmation before acting. This reduced total trade volume significantly but kept the subset of trades I did take at a more reasonable quality threshold.
Signal Quality in the Declining Regime
The declining regime presented a different set of problems from the volatile sideways regime. Volatility was elevated but less extreme than in the sideways phase — the KOSPI was moving directionally, just in the wrong direction. The more significant issue was the effect of persistent foreign and institutional selling on individual stock patterns.
During this period, I observed a specific failure mode that didn't exist in the other two regimes: what I started calling the "false morning star" pattern. A stock would show what looked like a clean wave formation and pullback in the first 30-40 minutes of the session, driven by overnight buying interest or technical positioning. The system would confirm the breakout. And then, typically between 10:30 and 11:30, the broader market pressure would overwhelm the individual stock's pattern — the stock would give back not just the breakout gain but the entire move from the morning low.
This pattern appeared in approximately 30% of confirmed signals during the declining regime — a rate high enough to significantly distort the overall statistics. The common characteristic of stocks that showed this pattern was that their initial move had occurred on retail-dominated buying with minimal institutional participation. The retail buyers were responding to the technical pattern in isolation; the institutional sellers, responding to the broader market direction, were ignoring it.
The adaptation here was more structural than just raising the quality threshold. During the declining regime, I added a specific filter: if the KOSPI was more than 1% below its prior close at the time of signal confirmation, I required institutional or foreign buying to be demonstrably present in the participant breakdown before acting. This filter eliminated approximately half of the signals in this period but sharply improved the quality of the remaining subset. The 12% of signals that made it through this filter had a continuation rate close to 55% — not as strong as the trending phase but well above the declining regime average.
The Lesson About Regime-Aware Trading
The core insight from watching the same system across three regimes is that a signal detection framework built on price and volume patterns is fundamentally an instrument for measuring demand relative to supply in local conditions. It works well when those local conditions are aligned with broader market dynamics — when the stocks generating the pattern are doing so because of genuine demand that the broader market is also experiencing. It works poorly when local pattern conditions are satisfied by noise or short-term positioning that runs against the broader flow.
The practical implication is that the system's rules should not be applied uniformly across all regimes. The same grade thresholds, the same ATR multipliers, and the same decision framework that make sense in a trending upward market are too permissive in a volatile sideways market and too blind to macro context in a declining market.
Building regime awareness into the system more formally — rather than adapting intuitively as I did during these periods — is one of the specific improvements on the roadmap for the system's next phase. The monthly review process I described in an earlier post provides the data needed to identify regime transitions. The next step is building explicit decision rules for how the system's operating parameters should shift in response to those transitions.
Today's Investing Insight — Regime Changes and the Limits of Historical Testing
One of the most consequential limitations of backtesting is that historical data tends to contain a mix of regimes, and a strategy that performs well across the full historical period may do so by performing exceptionally well in one regime and poorly in another — with the regimes weighted in the historical data in a way that doesn't reflect how they'll be weighted going forward. A system tested on a five-year period that was predominantly trending upward will show strong historical results even if it performs poorly in sideways or declining regimes, simply because the favorable regime dominated the test period. Walk-forward testing — where the system is tested on a series of out-of-sample periods rather than a single historical window — is one technique for detecting this problem. But the most direct approach is what this post describes: running the system in live conditions across multiple regimes and measuring its performance in each one separately, rather than treating the aggregate as a meaningful single number.
---
This post documents a personal journey of building and running an algorithmic trading system across multiple market environments and reflects personal experience and perspective. Past performance in any market regime does not predict future results in that regime or others, and all investment decisions and their outcomes are the sole responsibility of the investor.
Comments
Post a Comment