66-Day Quant Training, Days 4–10: SMA, ADX, Ichimoku, and Risk Management

The Seven-Day Learning Map

Day-by-Day Notes

Days 4–5: Starting with an SMA Trend Strategy

The Simple Moving Average (SMA) is straightforward: calculate the arithmetic mean of the past $N$ closing prices.

What We Learned in Practice

• Day 4: We used the IBKR API to extract historical data for the MAG7 stocks (MSFT, AAPL, TSLA, NVDA, META, GOOGL, AMZN) and searched for an SMA window for each ticker. The same parameter set did not transfer cleanly across all seven.
• Day 5: We converted a daily SMA 200 strategy into a 1-minute SMA 5 intraday script. The shorter timeframe made signals more responsive and increased false signals sharply.

Day 6: Describing Volatility with Bollinger Bands

Bollinger Bands consist of three lines:

• Middle Band: SMA (typically 20-day)
• Upper Band: SMA + $k \times \sigma$ (typically $k = 2$)
• Lower Band: SMA - $k \times \sigma$

Where $\sigma$ is the standard deviation of closing prices.

Backtest Results and Reflection

Using AAPL with a "buy at lower band, sell at upper band" mean-reversion logic over ten years:

• The grid search's best parameters (SMA=20, σ=3.0) produced a Sharpe 2.05× above the mean, an overfitting warning.
• The 3σ bandwidth was rarely touched (only 1 trade out-of-sample), meaning the strategy sat in cash most of the time — low drawdown but also low returns.
• AAPL exhibited a strong uptrend over 2016–2026; a pure mean-reversion strategy sold too early at the upper band, missing extended trend gains.

Day 7: Using MACD to Read Momentum

MACD stands for Moving Average Convergence Divergence. It measures momentum through the relationship between two exponential moving averages.

Advanced Technique: Histogram Slope Filter

Pure crossover signals are prone to whipsaws in ranging markets. Adding a histogram slope filter:

• Only allow buy entries when the histogram is rising
• Only allow sell entries when the histogram is falling

This confirmation layer removes some false signals and delays entry. That is the trade: more confirmation, less timeliness.

Day 8: ADX Measures Strength, Not Direction

ADX (Average Directional Index) does not generate direction by itself. It measures whether a trend exists and how strong that trend is.

How ADX Is Calculated

ADX isn't derived directly from price — it goes through multiple layers of processing. The full calculation chain breaks down into five steps:

Step 1: Directional Movement (+DM / -DM)

For each bar, compare the current high/low against the previous bar:

In plain terms: +DM captures the magnitude of "upward breakouts," while -DM captures "downward breakouts."

Step 2: True Range (TR)

TR measures each bar's actual volatility range, accounting for gap openings.

Step 3: Wilder's Smoothing

Apply Wilder's smoothed moving average (typically $N = 14$) to +DM, -DM, and TR separately:

Step 4: Directional Indicators (+DI / -DI) and DX

+DI represents the proportion of upward momentum relative to total volatility; -DI represents the downward proportion. DX measures the "divergence" between the two.

Step 5: ADX = Smoothed DX

The final ADX applies one more round of smoothing to DX, making the trend-strength reading more stable and less reactive to single-day spikes.

ADX Value Interpretation

ADX works alongside +DI (Positive Directional Indicator) and -DI (Negative Directional Indicator):

• Buy: ADX > 25 and +DI crosses above -DI
• Sell: ADX > 25 and -DI crosses above +DI

Day 9: Ichimoku Puts Several Readings on One Chart

Ichimoku Kinko Hyo displays trend direction, momentum strength, and support or resistance on one chart through five lines:

Compared to a Single MA Strategy

While SMA only tells you "where price is relative to the average," Ichimoku simultaneously provides:

1. Forward-looking perspective: The cloud extends into the future, letting you anticipate support and resistance zones
2. Multi-dimensional confirmation: Tenkan/Kijun crossover + cloud position + Chikou Span — triple confirmation reduces false signals
3. Built-in momentum reading: The distance between price and cloud indicates momentum strength; convergence signals weakening momentum

The tradeoff: signals are more conservative, trading frequency is lower, and opportunities in choppy markets are often missed.

Day 10: How a $1.88 Million Account Went to Zero

There was no code on Day 10. We read about a trader who ran a quantitative strategy called Imhotep. The account reached $1.88 million, then went to zero overnight after untested stop-loss code, absent monitoring, 100% capital deployment, and leverage combined in one liquidation event.

My Safety Net Framework

Drawing from The Trading Bible's core philosophy, I distilled three survival principles:

Principle 1: Drive Single-Trade Risk Toward Zero

• Maximum risk per trade ≤ 1% of total capital
• Calculate leverage based on actual exposure, not nominal position size

Principle 2: Maximize Your Survival in the Market

• Even after 20 consecutive stop-losses, the account retains 80%+ of its capital
• Separate "trading capital" from "reserve capital" — the reserve is never touched

Principle 3: Ensure Positive Expected Value

Where $W$ = win rate, $\bar{G}$ = average gain, $L$ = loss rate, $\bar{L}$ = average loss. Only strategies with positive expected value are worth executing.

Problems That Recurred Across Seven Days

In-Sample Winners Rarely Stay Winners Out of Sample

Every parameter optimization is a battle against overfitting. Across Day 6 (Bollinger Bands), Day 7 (MACD), Day 8 (ADX), and Day 9 (Ichimoku), we repeatedly confirmed the same truth:

The best in-sample parameters often lose their lead out of sample.

How to cope:

1. Always split data into in-sample (training) and out-of-sample (validation) periods
2. When reading Sharpe heatmaps, choose the "plateau" over the "peak" — a stable parameter region is more reliable than a single optimal point
3. When in-sample Sharpe dramatically exceeds out-of-sample, that's the overfitting red flag

The R.E.A.L. Backtesting Framework

All backtests followed the R.E.A.L. principles:

• Realistic: Include commissions ($0.005/share) and slippage (0.05%)
• Evidence-based: Validate phase by phase, never skip steps
• Avoiding Bias: Apply signal.shift(1) — execute on the next bar's open to prevent look-ahead bias
• Logged: Document observations in Markdown at every phase for reproducibility

Indicator Selection Quick Reference

Five Rules from the Week

1. Simple strategies tend to be more robust. The more complex the system, the more likely it breaks in live trading.
2. Backtests are maps, not territories. Historical data is perfect; markets are not.
3. Risk management comes before everything. Survive first, profit second.
4. Lagging isn't a flaw — it's a feature. An extra layer of confirmation means fewer false signals.
5. Don't chase "optimal" — chase "robust." Peaks on a Sharpe heatmap are traps; plateaus are where you want to be.