AI Model for Technical Indicators Analysis RSI MACD Bollinger

Development of AI-based Technical Indicators Analysis Model: RSI, MACD, Bollinger

Technical analysis has existed longer than modern neural networks. RSI, MACD, Bollinger Bands — empirically discovered tools that describe market dynamics. The ML approach does not replace these indicators, but automatically learns from them: finding optimal parameters, combinations and context of application.

RSI: Mean Reversion Prediction

Standard RSI(14):

RSI = 100 - 100 / (1 + RS)
RS = Average Gain / Average Loss over 14 periods

Traditional rule: RSI < 30 → oversold → buy signal. Problem: performs poorly in trending markets (RSI can stay below 30 for weeks in downtrend).

ML Enhancement:

• Period optimization (7-28) via Optuna for each instrument
• RSI in trend context: rsi_divergence_from_trend = rsi - trend_adjusted_rsi
• Multi-timeframe RSI: 7d, 14d, 21d as separate features
• RSI velocity: RSI change over 3 days (acceleration signal)

MACD: Momentum and Cross-Signals

Standard MACD:

MACD Line = EMA(12) - EMA(26)
Signal Line = EMA(9) of MACD Line
Histogram = MACD - Signal

ML Features from MACD:

• macd_histogram: value and sign
• macd_histogram_trend: increasing or decreasing (momentum of momentum)
• macd_signal_cross: 1 on bullish cross (MACD crossed Signal from below)
• macd_zero_cross: zero line crossing
• macd_divergence: new price low, but higher MACD — bullish divergence

Parametric Optimization: Standard (12, 26, 9) — not necessarily optimal for specific instrument/timeframe. Bayesian optimization of (fast_period, slow_period, signal_period) space by Sharpe ratio on walkforward validation.

Bollinger Bands: Volatility and Breakouts

Standard Bollinger Bands(20, 2):

Middle Band = SMA(20)
Upper Band = SMA(20) + 2 × σ(20)
Lower Band = SMA(20) - 2 × σ(20)

ML Features:

• %b = (Price - Lower) / (Upper - Lower): price position in band (0-1)
• bandwidth = (Upper - Lower) / Middle: band width, volatility proxy
• squeeze: bandwidth < 6-month minimum → breakout coming
• band_walk: price touches upper/lower band for several days → trending market

Bollinger Squeeze Strategy: Squeeze precedes significant movement. Direction uncertain, but magnitude expected. ML predicts breakout direction from squeeze.

Combined ML Model

Integration of all indicators into single model:

Feature matrix:

def build_technical_features(df):
    features = {}

    # RSI family
    for period in [7, 14, 21]:
        features[f'rsi_{period}'] = talib.RSI(df.close, period)
    features['rsi_divergence'] = compute_rsi_divergence(df)

    # MACD family
    macd, signal, hist = talib.MACD(df.close, 12, 26, 9)
    features['macd_hist'] = hist
    features['macd_hist_trend'] = hist.diff(3)
    features['macd_cross'] = (macd > signal).astype(int)

    # Bollinger
    upper, middle, lower = talib.BBANDS(df.close, 20, 2, 2)
    features['bb_pct_b'] = (df.close - lower) / (upper - lower)
    features['bb_bandwidth'] = (upper - lower) / middle
    features['bb_squeeze'] = (features['bb_bandwidth'] < features['bb_bandwidth'].rolling(126).min() * 1.1)

    # Stochastic
    slowk, slowd = talib.STOCH(df.high, df.low, df.close)
    features['stoch_k'] = slowk
    features['stoch_cross'] = (slowk > slowd).astype(int)

    return pd.DataFrame(features)

Model: LightGBM with quantile losses. Target: forward 5-day return. Feature importance will reveal which indicators are truly predictive.

Context-Dependency of Indicators

Key insight: indicators work differently in different market regimes:

• In trending market: RSI < 30 = continuation of decline (not reversal!)
• In ranging market: RSI < 30 = real reversal signal
• High volatility: Bollinger Bands need expansion (3σ instead of 2σ)

Regime-conditional model:

# Regime detector
regime = classify_market_regime(df)  # 'trending', 'ranging', 'volatile'

# Different models for different regimes
if regime == 'trending':
    signal = momentum_model.predict(features)
elif regime == 'ranging':
    signal = mean_reversion_model.predict(features)

Backtesting Technical Approach

Simple strategy for model validation:

1. When signal > threshold: long position N lots
2. When signal < -threshold: short
3. Stop: -2 × ATR(14)
4. Target: +4 × ATR(14)

Real transaction cost calculation:

• Commission: 0.04% per side (Interactive Brokers)
• Slippage: 0.02-0.05%
• Financing: overnight swap for forex/CFD

Evaluation: Sharpe > 1.0 on out-of-sample after TC — sufficient result for further development.

Timeline: baseline model with RSI+MACD+Bollinger features — 2-3 weeks. Full system with regime detection, parametric optimization and backtesting — 6-8 weeks.