PEG Strategy: Finding Undervalued Growth Stocks with Price/Earnings-to-Growth Ratio

Low P/E Doesn't Mean Cheap: Beware of Value Traps

Many investors get excited when they see a low P/E ratio, thinking they've found a bargain. But have you ever asked: Why is the market giving it such a low valuation?

The answer might be: the company is in decline.

A company with a P/E of just 5x isn't actually cheap if its revenue is shrinking and profits are declining year after year—the market is simply reflecting its dim future.

Conversely, a company with a P/E of 25x might be truly undervalued if its earnings are growing 50% annually.

This is why Peter Lynch introduced the PEG (Price/Earnings-to-Growth) ratio in his book "One Up On Wall Street":

PEG = P/E / Earnings Growth Rate

When PEG < 1, the stock price is cheap relative to its growth; PEG > 2 may indicate overvaluation.

Today we'll use data to verify: Does the PEG strategy really work in the Taiwan stock market? Through 4 rounds of iterative optimization, we found a strategy with 25.6% CAGR and only 37% maximum drawdown.

PEG Strategy Performance in Taiwan Stocks

Before diving into optimization, let's see how the basic PEG strategy performs.

Build a basic PEG strategy using FinLab: select the 10 stocks with the lowest PEG, rebalance monthly. Conditions: PEG between 0-2, positive operating profit growth, basic liquidity.

from finlab import data
from finlab.backtest import sim
 
# Get data
pe = data.get("price_earning_ratio:本益比")
營業利益成長率 = data.get("fundamental_features:營業利益成長率")
當月營收 = data.get("monthly_revenue:當月營收")
volume = data.get("price:成交股數")
 
# Calculate PEG
peg = pe / 營業利益成長率
 
# Build conditions
cond = (
    (peg > 0) & (peg < 2) &           # Reasonable PEG range
    (營業利益成長率 > 0) &             # Ensure positive growth
    (當月營收.average(3) / 當月營收.average(12) > 1.1) &  # Revenue momentum
    ((volume / 1000).average(20) > 300)  # Liquidity
)
 
# Select 10 stocks with lowest PEG
position = (peg * cond)
position = position[position > 0].is_smallest(10)
position = position.reindex(當月營收.index_str_to_date().index, method="ffill")
 
# Backtest
report = sim(position, resample="M", stop_loss=0.15, upload=False)
stats = report.get_stats()
print(f"CAGR: {stats['cagr']:.1%}")
print(f"Sharpe Ratio: {stats['monthly_sharpe']:.2f}")
print(f"Max Drawdown: {abs(stats['max_drawdown']):.1%}")

Results:

The basic strategy does beat savings deposits, but has two obvious problems:

1. Mediocre returns: 9.7% is only slightly better than the market
2. Excessive drawdown: 50% max drawdown is hard for most people to stomach

This is what we need to optimize.

AI-Assisted Iterative Optimization: From 9.7% to 25.6%

Here's the core of this article: the complete process of 4 rounds of AI-assisted optimization.

I set three optimization targets:

Iteration 1: Exploring Parameter Space

Test different parameter combinations: stricter PEG thresholds, adding RSI momentum, different holding sizes, different rebalancing frequencies. Find which factors have the most impact on performance.

# Test weekly rebalancing + RSI momentum filter
rsi = data.indicator("RSI", timeperiod=14)
roe = data.get("fundamental_features:ROE稅後")
 
cond_rsi = (
    (peg > 0) & (peg < 1.5) &
    (營業利益成長率 > 5) &
    (當月營收.average(3) / 當月營收.average(12) > 1.1) &
    ((volume / 1000).average(20) > 500) &
    (roe > 5) &
    (pe > 5) & (pe < 30) &
    (rsi > 50)  # RSI momentum upward
)
 
position = (peg * cond_rsi)
position = position[position > 0].is_smallest(10)
position = position.reindex(當月營收.index_str_to_date().index, method="ffill")
 
# Weekly rebalancing
report = sim(position, resample="W", stop_loss=0.10, upload=False)

Testing revealed an important phenomenon:

Finding: Increasing rebalancing frequency caused CAGR to jump from 9.2% to 28.8%! But max drawdown also reached 54.8%, still exceeding our target.

Iteration 2: Attempting to Reduce Drawdown

Try various risk control methods to reduce drawdown: stricter stop-loss, trailing stop, take-profit.

# Test different stop-loss/trailing stop configurations
configs = [
    {"stop_loss": 0.10, "trail_stop": None},   # Baseline
    {"stop_loss": 0.08, "trail_stop": None},   # Strict stop-loss
    {"stop_loss": 0.10, "trail_stop": 0.05},   # Trailing stop
    {"stop_loss": 0.10, "take_profit": 0.20},  # Take profit
]
 
for cfg in configs:
    report = sim(position, resample="W", **cfg, upload=False)
    # ... record results

Surprisingly:

Finding: Traditional stop-loss/trailing stop doesn't work for growth stock strategies! Because growth stocks fall together during systemic declines, stop-losses only lock in losses.

Iteration 3: Adjusting Strategy Structure

Since risk controls don't work, let's adjust the strategy structure itself.

Test different rebalancing frequencies (weekly, bi-weekly, monthly, quarterly) and holding sizes, observe impact on MDD.

# Test bi-weekly rebalancing
report = sim(position, resample="2W", stop_loss=0.12, upload=False)

Key Finding: Reducing rebalancing frequency effectively lowers MDD! Quarterly rebalancing has only 37% MDD while maintaining 25%+ returns.

Backtest Results

After 4 iterations, we found a strategy configuration that meets all targets.

Strategy Analysis: Why Does Quarterly Rebalancing Have the Lowest MDD?

This finding seems counterintuitive: reducing trading frequency actually reduces risk? Here's why:

1. Avoid chasing highs and selling lows: High-frequency rebalancing tends to buy at peaks and sell at bottoms
2. Reduce transaction costs: Frequent trading accumulates fees and slippage that erode returns
3. Let winners run: Good stocks need time to develop; selling too early misses the big moves
4. Filter short-term noise: Quarterly timeframe only considers medium-term trends, not daily fluctuations

Growth stock strategy MDD is hard to reduce with traditional risk controls—you must adjust the strategy structure itself.