Foreign Investors Keep Selling Taiwan Stocks? What Actually Works: Institutional Following, Backtested

Can You Really Make Money Going Against Foreign Investors?

A common saying online:

"Foreign investors are harvesting retail investors, just do the opposite to make money!"

Is this assumption reasonable? Let's verify with data (2015-01 to 2026-06, fees and transaction tax included):

• Reverse Foreign Investor Strategy: +8.8% CAGR — far behind 0050's +20.6% (with dividends)
• Pure Trust Fund Buying: +9.7% CAGR — also behind the index
• Trust Fund Buying + Revenue Momentum: +33.9% CAGR, 1.11 daily Sharpe

Following any single institutional flow signal, with or against it, loses to simply holding 0050. Only combining trust-fund buying with revenue momentum beats the index.

2026-06 data refresh: the figures originally cited in the January 2026 version (-11.2% / 17.5% / 31.68%) came from an older legacy backtest. All numbers on this page have been re-run with the exact code shown below on data through 2026-06-11; the headline conclusion is unchanged.

Hypothesis Testing: Pure Reverse Foreign Investor Strategy

First, let's test the original hypothesis: buy when foreign investors sell.

Backtest "Reverse Foreign Investor" strategy:

• Condition: Foreign investors sell for 5 consecutive days
• Buy stocks meeting the condition
• Hold for one month then rebalance

from finlab import data
from finlab.backtest import sim
 
# Get foreign investor trading data
外資買賣超 = data.get('institutional_investors_trading_summary:外陸資買賣超股數(不含外資自營商)')
vol = data.get('price:成交股數')
 
# Condition: Foreign investors sell for 5 consecutive days
外資連續賣 = (外資買賣超 < 0).sustain(5)
vol_filter = vol.average(20) > 300 * 1000
 
# Reverse operation: buy when foreign investors sell
position = 外資連續賣 & vol_filter
 
# Backtest
report = sim(position.loc['2015':], resample='M', position_limit=0.1, upload=False)
report.to_html('reverse_foreign.html')

Reverse Foreign Investor Strategy Performance

Reverse foreign investor strategy doesn't work: it takes index-level drawdowns while earning less than half the index return. Foreign investors may have valid reasons for selling (company declining), so treating their selling as a buy signal assumes they are always wrong.

Change of Thinking: Trust Funds More Predictive Than Foreign Investors?

Since reverse foreign investor doesn't work, what about following institutional buying?

We discovered a key insight: Trust funds have more predictive power than foreign investors

Reasons:

1. Focus: Trust funds focus on Taiwan stocks, foreign investors are global
2. Information Advantage: Trust funds research small-mid caps more deeply
3. Holding Period: Trust funds hold longer, indicating stronger conviction

Backtest "Trust Fund Buying" strategy:

• Condition: 10-day cumulative trust fund buying > 15,000 shares
• Liquidity: 20-day average volume > 300 lots
• Select top 10 by trust fund buying amount

from finlab import data
from finlab.backtest import sim
 
# Get trust fund trading data
投信買賣超 = data.get('institutional_investors_trading_summary:投信買賣超股數')
vol = data.get('price:成交股數')
 
# Trust fund cumulative buying
trust_sum = 投信買賣超.rolling(10).sum()
trust_buy = trust_sum > 15000
 
# Liquidity filter
vol_filter = vol.average(20) > 300 * 1000
 
# Select top 10 by trust fund buying
cond = trust_buy & vol_filter
position = trust_sum[cond].is_largest(10)
 
# Backtest
report = sim(position.loc['2015':], resample='M', upload=False)
report.to_html('trust_buy.html')

Pure trust fund buying achieves +9.7% annualized with a 0.44 daily Sharpe and a -45.5% max drawdown — slightly higher return than reverse foreign investor, but worse risk-adjusted, and still far behind 0050. Flow alone is not enough.

Problem Analysis: Pure Flow Isn't Enough

A problem with pure trust fund buying: buying stocks without fundamental support

Observations:

1. Some stocks with trust fund buying have declining revenue
2. Some are short-term speculative plays without sustainability

Solution: Add revenue screening to ensure improving fundamentals

The VIP section continues with the full institutional following strategy: 33.9% CAGR, Sharpe 1.11, plus strategy rules, the interactive backtest, and Python code.