ACADEMY / Strategy generation

Foreign Open Interest as a Signal: Mind the Release Time

Foreign futures open interest and short positions can be tested, but only as a delayed rule: the data is published after the close, so the earliest you can use it is that night's night session or the next day session. More than one cause can sit behind a contract count, and the history is short, so a clean-looking pattern picked from over a thousand candidates is mostly hindsight polish. Write the rule down first, then check it on out-of-sample data.

By Soren Huang · Published 2026-11-07 · updated 2026-11-07 · Editorial policy

Contents

Can foreign futures open interest and short positions work in a strategy?

You can test foreign futures open interest and short positions, but only as a delayed rule: the data comes out after the day's close, so the earliest you can act on it is that night's night session or the next day session. The usable history is only a few years, a contract count can have more than one cause behind it, and patterns that look clean are mostly picked after the fact. Write the rule down first and then backtest, and most patterns fade or disappear.

This article covers one thing: turning the foreign futures open interest that TAIFEX publishes from “a number in an after-close news item” into “a rule with correct timestamps that can be proven wrong.” My view is blunt: a count by itself is not a direction, and the best rule picked from a few years of data will always look better than reality. If you have never written a backtestable rule, read program trading for TAIEX futures from zero first to get rules, trading sessions and costs straight, then come back to foreign investor data.

Delayed position data is not unique to Taiwan. The US Commodity Futures Trading Commission (CFTC) Commitments of Traders report counts positions as of Tuesday's close and publishes them on Friday, with three trading days in between; any rule built on it can only act after Friday's close. TAIFEX publishes its institutional investor data every day, with a much shorter delay, but the point is exactly the same: you can use it only after it is published.

What does foreign futures open interest mean, and how is it different from net position?

Open interest and net position are two different numbers. The same net short of 20,000 contracts can mean few longs and many shorts, or large positions on both sides, and the meaning differs. Open interest is the number of long contracts and short contracts that a class of traders (for example, foreign and Mainland Chinese investors) has not yet closed on a futures contract, reported as two figures: long open interest and short open interest. Net position is long open interest minus short open interest; below zero is net short, above zero is net long. When the news says “foreign short positions,” it usually means the net position, not the total on the short side.

An everyday analogy: one company has 100 million in receivables and 80 million in payables, for net assets of 20 million; another has only 30 million in receivables and 10 million in payables, also for net assets of 20 million. The net is the same, but the first company's balance sheet is nothing like the second's. Futures positions work the same way, so in a backtest keep all three series, long, short and net, not just the net. For how a futures contract works and what one contract represents, see futures trading basics.

Hypothetical example

Scenario A: foreign investors hold about 40,000 long and about 60,000 short contracts in TAIEX futures, for a net position of −20,000 contracts. Scenario B: about 5,000 long and about 25,000 short, for the same net position of −20,000 contracts. After the close, the news reports both as “foreign investors net short 20,000 contracts.” But in Scenario A foreign investors hold large positions on both sides, and part of that may well be offsetting or split across different funds, while Scenario B is almost purely a bearish bet. The same rule, “go short when net short exceeds 15,000 contracts,” is really measuring different things in these two scenarios.

Both show a net short of 20,000 contracts. On the left is a hedged position that is large on both sides; on the right is a nearly one-sided short position. Looking only at the net mixes the two scenarios together.

Number	Definition	Use in a backtest
Long open interest	Number of long contracts not yet closed	Shows whether longs are adding or cutting
Short open interest	Number of short contracts not yet closed	Shows whether shorts are adding or cutting
Net position	Long open interest minus short open interest	A summary of direction, but hides the size of each side
Daily change in net position	Today's net position minus yesterday's	The series most often used as a signal
Contracts traded that day	Buy-side and sell-side contracts newly traded that day	Separates “new positions” from “closing out”

What does an increase in foreign net short positions mean?

A rise in foreign net shorts only says that foreign investors' short positions in futures have grown. At least three causes can sit behind it, and the count alone cannot tell them apart. A directional bearish bet is one, but foreign investors hold large Taiwan equity positions in the cash market, and using futures shorts to hedge those holdings is routine; around settlement week, closing the near month and opening the far month can also make the shorts in one contract temporarily rise. All three look like “net shorts +5,000 contracts,” yet they mean different things for the next day's market.

Directional bearish bet (shorting outright). To tell this one, the same day's foreign cash-market trading should also be net selling, so futures and cash point the same way.
Cash hedge (holding Taiwan stocks and hedging with futures shorts). The typical sign is net buying in cash, or unchanged holdings, while futures shorts increase; foreign investors' overall exposure has not actually turned short.
Roll or arbitrage (a temporary gap from closing the near month and opening the far month). This is the most common misread when you look only at the near-month contract, because positions the far month has already taken over look as if they vanished or appeared in the near-month column. TAIFEX data can be split by contract month; around settlement week, add up all months before you judge.

The same “net shorts up 5,000 contracts” has at least three possible sources. To tell them apart you need cash-market net buying and selling and the data by contract month; one number alone cannot do it.

This is why “how many contracts count as a lot for foreign investors” has no fixed answer. Foreign holdings of Taiwan stocks, the index level and total open interest all change, so 20,000 contracts represented a different share of exposure ten years ago than it does now. The more robust approach in a backtest is to use relative measures instead of absolute counts, for example net position divided by total market open interest, or the daily change in net position relative to its average change over a past period. That way the same rule measures the same thing in different years.

Where to find foreign open interest: TAIFEX institutional futures and options data

TAIFEX publishes daily the trading volume and open interest of the three major institutional investors (foreign and Mainland Chinese investors, investment trusts, dealers) in futures and options. You can look them up by date, product and investor type, and the first step before backtesting is to organize them into a time series with one row per day. The data sits under “Trading Information” on the TAIFEX website, with a set of tables each for futures and options. For futures you can choose “by date” or “by futures contract”; only the latter shows the figures for individual products such as TAIEX futures (TX) and Mini-TAIEX (MXF), and only it lets you split by contract month.

Decide which fields to pull. At a minimum: date, foreign and Mainland Chinese investors' long open interest and short open interest in TAIEX futures, and long and short trading volume. Add investment trusts, dealers, and options call and put positions if you need them.
Decide the contract scope. Whether to pull only the near month or the total of all months, settle it at this step and write it in your notes. Around settlement week, the two choices give different numbers.
Arrange one row per day. One trading day per row, fixed columns, one date format. There is no data on days when TAIFEX is closed, and those days should not have a daily bar in your price data either.
Compute the derived fields yourself. Recalculate net position, daily change in net position and the ratio to total market open interest from the raw fields; do not copy figures from news sites or third-party summaries.
Set aside a recent held-out period. Do not load the last few months into any research; once the rule is final, test on them only once.

Historical data on the TAIFEX site goes back many years, but the history you can actually use in a backtest has to exclude the stretches where market structure changed: foreign investors' behavior in futures may differ before and after the night session launched, and before and after Micro TAIEX futures were introduced. Each year has fewer than 300 trading days of observations, so ten years is only a little over 2000 rows, and that count becomes a constraint in the next section and the last one.

Release is delayed: how do you write a backtestable rule with correct timestamps?

Institutional investor data is published after the day session closes, so a rule can only use “yesterday's published numbers” to decide “today's action,” and you must also decide in advance which bar to trade, in that night's session or in the next day session; otherwise you have look-ahead bias. Look-ahead bias is when a backtest uses, on some bar, information that did not yet exist; the result looks good but cannot be done live. Take foreign investor data dated T: it describes positions at the close on day T and is also published after the close on day T. Any approach that maps it to the open on day T, or to any bar of the day T day session, trades the past with future data. The common leaks are collected in look-ahead bias in backtests and the one-bar check.

The top row maps day T foreign investor data to the day T open, which is look-ahead bias. The bottom row maps it to the first usable bar after release, the only alignment you could trade live.

There are two ways to pick the first usable bar, and each has a cost. The TAIEX futures day session currently runs 08:45 to 13:45 and the night session 15:00 to 05:00 the next day (TAIFEX may change the hours, so go by the contract specifications page on the TAIFEX site; you can also check with the free futures market hours tool). Choosing the first bar of that night's session has the advantage of being closest to the data release; the drawbacks are that the night session is thinner, so you should set slippage higher, and that data vendors do not agree on which trading day a night-session bar belongs to; full-session futures backtests and how to split trading days explains how to make it consistent. Choosing the first bar of the next day session has the advantage of good liquidity and a clear trading-day definition; the drawback is that a whole night session sits in between, and moves in US stocks are already priced into the open. If data published on Friday is mapped to the Monday day session, a whole weekend sits in between as well, and how you handle Monday open gaps in futures backtests will affect this rule's results.

Five steps from the TAIFEX raw table to a backtestable rule. The date shift in step 4 is the step most easily missed in the whole process, and the one most likely to distort the results.

Hypothetical example

A correctly written rule: “If the daily change in foreign investors' TAIEX futures net position published for day T is below −5,000 contracts, and foreign investors were also net sellers in the cash market the same day, then short one contract at the open of the first five-minute bar of the day T+1 session and exit at the close of the day T+1 session; commission and one tick of slippage are charged on every trade.” The entry condition, order bar, exit time and costs are all fixed before the backtest runs; afterward you can only read the results, not go back and change the threshold.

Which patterns in a few years of data are just coincidence? What foreign investor data cannot show

The usable history of foreign investor data is only a few years to a dozen or so, with fewer than 300 observations a year; pick the best rule out of over a thousand candidates and the result is bound to be flattered. The open interest count also cannot see the cash positions that hedges offset, or intraday entries and exits. This is not unique to foreign investor data; it is a problem for any short history combined with many candidates. Bailey et al. (2014) show that the more strategy variants you try, the more the best one's in-sample performance exceeds its true performance on new data; Harvey and Liu (2015) propose discounting results by the number of trials. An everyday analogy: have a thousand people each flip a coin ten times and someone will get nine heads, but you would not say that person is good at flipping coins.

The results of over a thousand candidates mostly cluster near zero, and the one picked afterward lands in the right-hand tail. Its figure reflects the luckiest candidate; what you can realistically expect sits near the middle.

The backtest below makes this “selection effect” concrete. It runs on ES (CME E-mini S&P 500 futures) daily bars from 2019-01-02 to 2025-06-30, using a single entry and exit rule written with RSI, with no stop loss, no profit target and no filter. It is a stand-in rule built on a standard indicator, not the foreign open interest itself; the aim is to demonstrate the process, not to test this article's subject. This rule is the one kept, ranked by return relative to drawdown, from the 1,024 candidates tried in the search. When you put candidates like these through AlphaPrime (Valynox) strategy validation, write the out-of-sample and walk-forward thresholds first, so that a candidate that fails an earlier check never reaches the later data; far fewer candidates then get to the final stretch of data.

Source: AlphaPrime product engine (CPU), run 2026-10-31. ES daily, 2019-01-02 to 2025-06-30, long and short, 1 contract, commission $20 and slippage $25 per trade.

Number	Result
Net profit (closed trades)	$89,155
Max drawdown	$68,387.5
Trades	56
Win rate	58.93%
Profit factor	1.693
Average trade	$1,592.05
Time in market	92.09%

Equity curve and drawdown of the candidate that was kept. Look at which years the slope flattens, how long the deepest drawdown lasts, and whether the section marked out-of-sample has a shape similar to the earlier part. ES · daily · 2019-01-02 to 2025-06-30 · CPU engine backtest

How to read this box: net profit of $89,155 against a maximum drawdown of $68,387.5 means that over six and a half years the rule fell back from a peak by an amount close to the total it eventually earned; 56 trades with 92.09% time in market mean it held a position almost the whole time, so most of the result comes from how ES itself moved over this period. More important, it was picked after the fact from 1,024 candidates, so these figures are bound to look better than the same rule's figures on new data. You cannot say RSI made this money, and you cannot treat it as a result anyone can expect; it only shows what “picking the best one” looks like. Put a foreign open interest rule through the same process, with as many candidates, a shorter history and fewer observations per year, and the flattering effect can only be larger.

Beyond the selection effect, foreign futures data itself cannot show three things:

The cash positions behind a hedge. The data does not say which holdings the futures shorts hedge or at what hedge ratio; you can only infer it indirectly from cash-market net buying and selling.
Intraday entries and exits. What is published is a snapshot at the close; positions foreign investors open and close during the session are not in it at all.
Disagreement within foreign investors. “Foreign and Mainland Chinese investors” is one combined category, containing bulls, bears and pure arbitrageurs, and the net cancels all of them out.

As a next step: download from TAIFEX the daily open interest of foreign and Mainland Chinese investors in TAIEX futures for the last few years, arrange it one row per day as in the fourth section, shift the dates one day later to line up with your bars, write one complete rule like the example in the fifth section, count how many trades it would produce in-sample, and only then decide whether it is worth the time to run it.

Frequently asked questions

How many contracts count as a lot for foreign investors?

There is no fixed contract threshold. Foreign investors' holdings of Taiwan stocks, the index level and total market open interest all change over time, so 20,000 contracts represent a different share of exposure in different years. In a backtest, use relative measures, for example net position divided by total market open interest, or the daily change in net position relative to its average change over a past period; then the same rule measures the same thing in different years.

How is foreign open interest different from the retail long-short ratio?

Foreign open interest is the long and short open interest of foreign and Mainland Chinese investors in futures as published by TAIFEX, taken from its statistics by investor type. The retail long-short ratio is usually derived by subtracting the three major institutional investors' positions from Mini-TAIEX (MXF) open interest, so it is an estimate, not a directly published statistic. Both are published after the close with a delay, and aligning them with bars works the same way. For how to write the Mini-TAIEX retail long-short ratio as a rule, see testing the Mini-TAIEX retail long-short ratio as a signal.

What do I do when foreign shorts suddenly rise or fall in settlement week?

First add up the positions across all contract months of the same product before looking, because closing the near month and opening the far month can cause a large temporary swing in a single month's figure. You can also mark the few trading days around settlement in your rule and tally their results separately. You can check TAIEX futures settlement dates with the free TAIEX futures settlement dates tool.

Why does a pattern in foreign investor data look good in a backtest but get worse in live trading?

The two most common causes are look-ahead bias and the selection effect. The first maps day T data to a day T bar, using numbers that were not yet published; the second picks the best of many candidate rules, whose in-sample result is bound to look better than its result on new data. They are handled, respectively, by a correct date shift and by out-of-sample thresholds written beforehand. The research behind this is collected in backtest overfitting: why the best backtest is usually inflated.

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.
Harvey, C. R., & Liu, Y. (2015). Backtesting. The Journal of Portfolio Management, 42(1), 13–28.
U.S. Commodity Futures Trading Commission. Explanatory Notes: Commitments of Traders. https://www.cftc.gov/MarketReports/CommitmentsofTraders/ExplanatoryNotes/index.htm
Taiwan Futures Exchange. Mini-TAIEX Futures. https://www.taifex.com.tw/enl/eng2/mTX