Every trading session starts with the same problem: the Korean stock market has roughly 2,500 listed stocks. Monitoring all of them simultaneously is impossible. Picking randomly is worse than useless. The quality of what you end up trading on any given day is largely determined by the quality of your stock selection process — what gets onto your watchlist, and why.
After more than a year of running the signal detection system described in this series, I've developed a multi-stage filtering process for building the daily watchlist that the system monitors. This post describes that process in full — the specific filters I apply, in the order I apply them, and what each one is designed to eliminate.
Why Most Watchlists Don't Work
The most common approach to building a trading watchlist is to add stocks that have recently made news, stocks that have moved significantly in prior sessions, or stocks that feel interesting based on a general market narrative. This approach has two problems.
First, by the time a stock is in the news or has made a notable move, much of the opportunity has often already passed. The traders who benefited most from that stock's move were the ones who identified it before it became obvious. A watchlist built on recent headlines is systematically late.
Second, a watchlist built on narrative is vulnerable to confirmation bias. Once you've decided a stock is interesting because of a story you find compelling, you tend to see confirming signals and discount contradictory ones. The decision to add a stock to the watchlist contaminates the subsequent evaluation of whether to trade it.
My approach is deliberately mechanical: each filter eliminates stocks that fail a specific quantitative criterion, and the process produces a watchlist whose contents are determined by data rather than narrative.
Stage One: Universe Definition
The first stage defines the tradeable universe — the set of stocks that are even eligible for the watchlist. I exclude stocks that fail any of these criteria before any further analysis.
Minimum trading value: the stock must have averaged at least 5 billion KRW in daily trading value over the prior ten sessions. This eliminates illiquid stocks where the bid-ask spread makes precise entry and exit impractical and where large orders can move the price significantly.
Price range: I exclude stocks priced below 3,000 KRW. Very low-priced stocks often have wide spreads relative to price and exhibit more erratic behavior than higher-priced stocks. I also exclude stocks priced above 500,000 KRW, where position sizing becomes awkward for the capital levels I'm working with.
No recent trading suspensions: stocks that have been suspended from trading in the prior twenty sessions are excluded. Recent suspensions are often associated with disclosure issues or unusual corporate situations that create unpredictable price behavior.
No same-day circuit breaker violations: stocks that have triggered a daily price limit halt (the 30% up or down limit in Korean markets) on the current day are excluded. These situations are typically driven by extraordinary events that make normal pattern-based analysis unreliable.
After applying these filters, the eligible universe typically narrows from roughly 2,500 stocks to somewhere between 400 and 700, depending on market conditions.
Stage Two: Momentum Screen
From the eligible universe, the second stage identifies stocks showing the kind of price and volume behavior that the signal system is designed to track. I'm looking for stocks that have shown a meaningful move upward on elevated volume within the prior two sessions.
The specific criteria: the stock must have closed at least 4% above its five-session average closing price within the last two sessions, and its trading value on the day of that move must have been at least twice its ten-session average trading value. This combination — significant price move accompanied by significantly elevated volume — identifies stocks where something meaningful happened recently that may be producing a tradeable pattern.
This stage is not looking for stocks to trade directly. It's looking for stocks that have shown the kind of activity that sometimes leads to the wave-pullback-breakout pattern within the following session or two. Many of these stocks will not produce a signal — they'll either continue moving without pulling back, or they'll pull back too deep, or the pattern simply won't materialize. The momentum screen is the input funnel, not the output.
After this stage, the watchlist typically contains between 15 and 40 stocks on an average session, depending on how active the market has been in prior days.
Stage Three: Participant Flow Filter
The third stage is the one that has improved the quality of the final watchlist the most. For each stock that passes the first two stages, I check the investor participant breakdown over the prior three sessions: how much of the recent trading value has been driven by institutional investors, foreign investors, and retail investors respectively.
Stocks where the recent elevated volume is predominantly retail-driven move to a lower-priority category. I don't exclude them entirely — retail-driven momentum can still produce tradeable patterns — but they receive less attention during the session and require a higher signal grade before I act.
Stocks where institutional or foreign buying is a meaningful component of the recent elevated volume move to a higher-priority category. These stocks have shown that large-capital participants have been active in the recent move, which increases the probability that the buying interest is structural rather than episodic. They receive more monitoring attention during the session.
In practice, this stage sorts the watchlist into two tiers rather than filtering further. Tier one (institutional/foreign participation) typically contains 8 to 15 stocks. Tier two (retail-dominated) typically contains 10 to 25 stocks.
Stage Four: Chart Structure Check
The fourth stage is the only qualitative stage in the process. For each stock in tier one — the higher-priority group — I spend about two minutes reviewing the recent price chart.
What I'm looking for is whether the stock's recent price action shows a coherent structure: a clear move upward, some evidence of consolidation or pullback, and a setup that looks like it could produce a breakout in the current or near-term session. I'm not trying to predict whether the breakout will happen — that's what the signal system is for. I'm trying to eliminate stocks where the chart is so disorganized that even if a technical signal fires, it won't be meaningful in context.
Stocks that show messy, choppy price action with no identifiable structure get moved to tier two or removed from the watchlist entirely. Stocks that show a clean, coherent structure — even if no specific pattern is yet confirmed — remain in tier one.
After this stage, the tier-one watchlist typically contains 5 to 12 stocks. These are the stocks I monitor most closely during the session, and they're the primary candidates for acting on confirmed signals.
How the Signal System Interacts With the Watchlist
One important distinction: the signal system's input is defined by the conditional search (the screener set up in the brokerage platform), not solely by the manual watchlist. The system automatically adds stocks that meet the conditional search criteria — primarily trading value and price movement thresholds — to its monitoring queue during the session.
The manual watchlist serves a different purpose: it provides context for evaluating signals that fire during the session. When the signal system generates a confirmed signal, the first thing I check is whether the stock is in tier one, tier two, or not on the watchlist at all. A signal on a tier-one stock receives full attention and is evaluated with the expectation of acting if the other criteria are met. A signal on a stock that isn't on the watchlist at all receives more skepticism — I check why it didn't make the watchlist and whether the reason remains valid.
This interaction between the pre-session manual process and the session's automated detection produces better decisions than either alone. The automated detection is fast and comprehensive; the manual pre-session work provides the context needed to evaluate what the automated detection finds.
Today's Investing Insight — The Difference Between a Screener and a Watchlist
Screeners and watchlists serve related but distinct purposes. A screener applies quantitative criteria to a large universe of stocks to produce a list of candidates — it's designed for breadth and speed. A watchlist is a curated set of stocks you intend to monitor actively during a session — it's designed for focus and depth. Many traders conflate the two, treating screener output as a watchlist. The problem with this approach is that screener output is often too large to monitor meaningfully and lacks the contextual understanding that comes from having done specific research on each stock. A better workflow is to use a screener as the input to watchlist construction — exactly the process described in this post — rather than treating screener output as the final product. The goal of the watchlist is a small enough set of stocks that you can pay genuine attention to each one, with enough contextual understanding of each to interpret the signals that occur during the session.
---
This post documents a personal journey of building and running an algorithmic trading system and reflects personal experience and perspective. Stock screening and watchlist criteria described here are based on personal observation and are not a recommendation for any specific stocks or approach. All investment decisions and their outcomes are the sole responsibility of the investor.
Comments
Post a Comment