Mini-TAIEX retail long-short ratio: how to test it as a signal

Contents

1. What is the mini-TAIEX retail long-short ratio?
2. How is the ratio calculated from open interest?
3. When is the ratio published, and where does look-ahead bias come from?
4. How do you test a contrarian rule on the ratio?
5. Does the ratio add anything to a price rule?
6. Why can the ratio stop working?
7. Frequently asked questions
8. References

The mini-TAIEX retail long-short ratio (小台散戶多空比) is the net position of everyone outside the three institutional investor groups, divided by total open interest. It is computed from TAIFEX's daily open interest data after the close. It can be tested as a contrarian signal, but only if each value is used after it was published.

What is the mini-TAIEX retail long-short ratio?

The retail long-short ratio estimates whether small traders, as a group, hold more long or more short mini-TAIEX contracts. Mini-TAIEX futures (小台, contract code MTX) are TAIFEX index futures worth NT$50 per point of the TAIEX, a quarter of the standard TAIEX futures contract.

A contrarian indicator is a measure you trade against: when the group it describes leans heavily one way, you take the other side. The idea behind the 散戶多空比 is that small traders as a group are often on the wrong side near turning points. That is a belief. A backtest is how you check whether it held in the data, and how strongly.

How is the ratio calculated from open interest?

You subtract the institutional positions from total open interest and divide the retail net position by total open interest. Open interest is the number of contracts that are still open at the end of a trading day. Every open contract has one buyer and one seller, so total long open interest always equals total short open interest.

Each trading day, TAIFEX publishes the long and short open interest of three institutional groups for each futures contract. The groups are dealers (自營商), investment trusts (投信) and foreign investors (外資). The common calculation follows four steps:

1. Take the total open interest of mini-TAIEX futures across all delivery months.
2. Retail long = total open interest minus the long open interest of the three groups combined.
3. Retail short = total open interest minus the short open interest of the three groups combined.
4. Divide the difference between retail long and retail short by total open interest.

As one formula: ratio = (L − S) ÷ OI. Here L is the retail long contracts, S is the retail short contracts, and OI is the total open interest. The result is usually shown as a percentage. A positive value means the leftover group is net long. A negative value means it is net short.

The mirror image in the example is a key property. The ratio is the institutional net position with the sign reversed, scaled by open interest. When you test it, you are partly testing institutional positioning.

Definitions also vary between sources. Some sites also subtract large-trader positions. Some round differently, and some show a plain number instead of a percentage. That is why a 小台散戶多空比查詢 page may not match your own figure. For a backtest, pick one definition, write it down, and compute the whole history yourself from TAIFEX's raw files. Then every day in the test uses the same definition.

When is the ratio published, and where does look-ahead bias come from?

The ratio for a trading day exists only after that day's regular session has closed, so no bar of that session may use it. Look-ahead bias is a backtest error in which a rule uses information that was not yet available when the trade would have been made.

TAIFEX index futures trade in a regular session from 08:45 to 13:45 Taiwan time. They also trade in an after-hours session from 15:00 to 05:00 the next day. TAIFEX counts after-hours trades toward the next trading day. The open interest for day T is therefore known only after T's regular session ends.

The most common error is a date join. A daily file stamps the value with date T. A script then attaches it to every bar of date T, including the 08:45 bar. The morning bars now "know" a number that was published in the afternoon. The same error appears on daily bars when a rule reads T's ratio and fills at T's close. Use these rules instead:

1. Store the time each value became available next to the value, not only the date.
2. Let T's value act from the first bar after that time. If you do not know the exact time, use it from the next regular session open at 08:45.
3. If your bars include the after-hours session, decide whether the rule may act there and keep that choice fixed. The guide to day-only versus full-session bars in night-session backtests covers this choice.
4. Print a few days of bars with the attached ratio and check by eye that no bar shows a value before its publication.

Other TAIEX-specific settings, such as contract rolls and session boundaries, are listed in the five-setting TAIEX futures backtest checklist.

How do you test a contrarian rule on the ratio?

Define "extreme" relative to recent history, trade at the next session, include costs, and judge the rule on data its thresholds were not chosen on. A fixed level such as +30% is risky because the typical range of the ratio drifts over the years. A percentile adapts to that drift. A percentile is the value below which a given share of past values lies. For example, the 90th percentile of the last year is higher than 90% of that year's readings.

1. Signal. Short when the ratio is above its rolling 90th percentile. Go long when it is below its rolling 10th percentile.
2. Entry. Enter at the first bar after publication, or at the next 08:45 open.
3. Exit. Hold a fixed number of sessions, or exit when the ratio returns to its median. Choose one rule before looking at results.
4. Costs. Include commission, tax and slippage on every round trip. The guide to commission, slippage and the average trade in backtests shows how to compare cost with edge.
5. Trade count. Count separate trades, not signal days. Extreme readings cluster, so many signal days can become a handful of trades.
6. Out-of-sample check. Choose the percentiles and holding period on one part of the history and test them unchanged on the next part. Walk-forward analysis repeats this across the whole history.

Does the ratio add anything to a price rule?

The ratio is only useful if it improves a price rule you could trade without it, so test it as an addition and compare it with that baseline. The ratio is tied to price movement. If institutions add short positions during a fall, the retail net long figure rises at the same time. A contrarian rule on the ratio can then be a price rule in disguise, such as "short after the index has fallen."

Role of the ratio	Example rule	What to check
Filter	A trend rule takes longs only when retail is net short	Does the filtered rule beat the unfiltered rule out of sample, with enough trades left?
Trigger with confirmation	Ratio extreme, then price breaks the prior session's high or low	Does the confirmation alone, without the ratio, do just as well?
Veto	Skip new longs when retail is very net long	Do the skipped trades lose money on average, or are they ordinary trades?

The baseline test is simple. Build the closest price-only rule, run it on the same days and with the same costs, and compare. If the price-only rule captures most of the result, the ratio added little. AlphaPrime (Valynox) strategy generation turns price ideas into testable rules with parameter ranges and entry and exit conditions. That makes it a practical place to build these baselines.

Every extra choice, such as the percentile, lookback, holding period or confirmation rule, is another test. If you try many combinations and keep the best, the winner looks good partly by luck. This data-snooping problem is described by White (2000), and its effect on out-of-sample results by Bailey, Borwein, López de Prado and Zhu (2014). Harvey and Liu (2015) argue that backtest results should be discounted for the number of tests behind them.

Why can the ratio stop working?

The ratio is a daily, delayed and indirect measure, and the population behind it changes. A relationship found in one period may therefore weaken or reverse in another. These are the main reasons, with the matching check for each:

• One value per day, after the move. By the next open, the market has already reacted to the session that produced the reading. Measure results from the realistic entry, not from the close that created the signal.
• A leftover, not a survey. The retail group includes every account outside the three institutional groups. A shift in who trades under those labels changes the ratio without any change in small-trader opinion.
• A changing population. TAIFEX launched micro TAIEX futures in 2024. Some small traders may hold that contract instead of mini-TAIEX. Test the period after the launch separately.
• Expiry effects. TAIEX futures settle on the third Wednesday of the delivery month. Positions close and roll around that date, and open interest and the ratio can jump for mechanical reasons. Check whether many of your signals fall in expiry week.
• Regime changes. In strong trends, the leftover group can stay heavily positioned for weeks. Split the history into several sub-periods and check that the relationship has the same sign in each.

What this means if you use AlphaPrime: treat the ratio as an outside input that has to beat a price-only rule. Build that price rule in AlphaPrime and judge both with the same out-of-sample standards, such as the walk-forward analysis and thresholds in strategy validation.

Frequently asked questions

Where can I look up the mini-TAIEX retail long-short ratio?

TAIFEX publishes the inputs every trading day: the open interest of the three institutional groups by contract, and the total open interest. Many sites publish a computed ratio, but definitions differ. For a backtest, compute the history yourself from the TAIFEX data so every day uses the same formula.

Is a high retail long-short ratio bearish?

A high value means the non-institutional group holds more long than short mini-TAIEX contracts. The contrarian reading treats that as bearish, but this is a hypothesis to test, not a rule. Because the ratio mirrors the institutional net position, a high value also means institutions are net short.

Why does my calculation differ from a website's number?

Sources differ in whether they subtract large-trader positions, which delivery months they include, and how they round or scale the result. Write down your own definition and apply it to the whole history. Consistency matters more than matching any one site.

Can the retail long-short ratio be used for day trading?

There is only one value per trading day, and it becomes available after the regular session closes. A day-trading rule can only use the previous day's value as a background filter. It cannot react to changes within the session.

Does the mini-TAIEX ratio include micro TAIEX positions?

No. The mini-TAIEX ratio uses mini-TAIEX open interest only, because each contract's positions are reported separately. If small traders move between contracts, the group the mini-TAIEX ratio describes changes, so test periods before and after such shifts separately.

References

• Bailey, D. H., Borwein, J., López de Prado, M., & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society, 61(5), 458–471.
• White, H. (2000). A Reality Check for Data Snooping. Econometrica, 68(5), 1097–1126.
• Harvey, C. R., & Liu, Y. (2015). Backtesting. The Journal of Portfolio Management, 42(1), 13–28.

What AlphaPrime does here

AlphaPrime's strategy generation builds the price rules a ratio filter would sit on, and strategy validation applies walk-forward analysis and custom thresholds so every candidate is judged by the same standard.

See plans and pricing