Skip to content

Tutorial 3: Advanced Usage of PyTechnicalIndicators

In this tutorial we go deep on just one indicator: the Relative Strength Index (RSI).
We will systematically compare different constant_model_type variants and periods, score their signals, and rank configurations.
You can later apply the exact same framework to other indicators (see “Extending Further”).

Series so far:

🎯 Goal

You will learn how to:

  1. Enumerate RSI variants via different constant_model_type values.
  2. Iterate across a period grid (e.g., 5–30).
  3. Generate oversold / overbought signals (long/short candidates).
  4. Score each signal using a forward-looking heuristic.
  5. Aggregate scores into a ranked table of configurations.
  6. Plan how to extend the same pattern to other indicators.

🧩 Why Focus on RSI?

Keeping scope narrow:

  • Makes the scoring logic clearer.
  • Prevents tutorial bloat.
  • Allows you to internalize the evaluation pattern before generalizing.

🔧 Variant Dimensions

We will vary:

Dimension Values (Example)
constant_model_type simple, smoothed, exponential, median, mode

📂 Data Setup

import pandas as pd
import pytechnicalindicators as pti

df = pd.read_csv("prices.csv", parse_dates=["Date"]).sort_values("Date").reset_index(drop=True)
df = df.dropna(subset=["Open","High","Low","Close"])
close = df["Close"].astype(float).tolist()

🔧 Define Model Types and Period

CONSTANT_MODELS = [
    "simple",
    "smoothed",
    "exponential",
    "median",
    "mode"
]
RSI_PERIOD = 5
OVERSOLD = 30.0

🧮 Compute RSI Variants and Rate Signals

results = []

for ctype in CONSTANT_MODELS:
    # Compute RSI for this model type
    rsi_vals = pti.momentum_indicators.bulk.relative_strength_index(
        close,
        constant_model_type=ctype,
        period=RSI_PERIOD
    )
    # The output is aligned to the tail, so index mapping:
    start_idx = len(close) - len(rsi_vals)
    rating = 0
    total_signals = 0
    for offset, rsi in enumerate(rsi_vals[:-1]):  # skip last (no future bar)
        idx = start_idx + offset
        price = close[idx]
        if rsi < OVERSOLD:
            total_signals += 1
            next_price = close[idx + 1]
            if next_price > price:
                rating += 1
    success_rate = rating / total_signals if total_signals > 0 else 0.0
    results.append({
        "model": ctype,
        "period": RSI_PERIOD,
        "signals": total_signals,
        "correct_signals": rating,
        "success_rate": success_rate
    })

📊 Present the Results

import pandas as pd

score_df = pd.DataFrame(results)
score_df = score_df.sort_values("success_rate", ascending=False)

print("RSI Model Ratings (RSI < 30, period=5):")
print(score_df[["model", "period", "signals", "correct_signals", "success_rate"]])

best = score_df.iloc[0]
print(f"\nBest model: {best['model']} (Success Rate: {best['success_rate']:.2%})")

Example output:

RSI Model Ratings (RSI < 30, period=5):
         model  period  signals  correct_signals  success_rate
3       median       5       43               24      0.558140
0       simple       5       38               20      0.526316
1     smoothed       5       41               21      0.512195
4         mode       5       41               21      0.512195
2  exponential       5       46               22      0.478261

Best model: median (Success Rate: 55.81%)

🧠 Interpreting Results

Column Description
model The constant model type tested
period RSI period used (fixed here at 5 for comparison)
signals Number of "buy" signals (RSI < 30) generated
correct_signals Number of times the next price was higher (win)
success_rate correct_signals / signals
  • A higher success rate suggests that the model/period pair gives better "next bar up" predictions when oversold.
  • If signals is very low, results may not be reliable—try longer history or adjust period.

📝 Extending Further

  • Try other periods (repeat the loop for different RSI_PERIOD values).
  • Add "sell" signal evaluation (RSI > 70 and next bar down).
  • Apply similar logic to other indicators (e.g., bands, moving averages).
  • Try longer-forward horizons (e.g., check next N bars for improvement).
  • Visualize where signals occurred on the price chart (see Tutorial 2).

🧩 Putting It All Together

A runnable example of the full code can be found in 03_advanced_pytechnicalindicators.py

🛡️ Disclaimer

This is a didactic example, not a trading strategy.
Success rate on "next bar" is just one toy metric—real research requires out-of-sample testing, risk modeling, and more.

✅ Next Step

Continue to: 04 - Connecting to an API

Happy model testing! 🦀🐍📈