Devlog #1

Welcome to the development log! This project tracks our engineering journey from a flawed, overfitted stock picker into an adaptive, blind multi-asset allocation system. Think of it like building a self-driving car, but for managing money.

The Development Journey

Step 1: The “Guessing” Car (Overfitting)

We started by creating a basic computer rule:

“Buy stocks when they drop a bit, and sell them when they jump.” We let the computer guess the absolute best numbers for this rule using past stock market data. The computer came back with perfect numbers that looked like a jackpot, but there was a catch: it only worked perfectly on the past data because it memorized the answers. If we tried to use this exact code in the real world, it would crash instantly.
So I had to:

Identify initial strategy flaws
Diagnose curve-fitting problems

Step 2: The “Adaptive” Car (Walk-Forward Analysis)

To fix the guessing problem, we forced the computer to study history in small chunks.

The bot looks at a 10-year window of data to learn basic market behaviors.
It tests its rules on the next 5-year window to prove it can adapt to unseen data.

This forced the computer to dynamically adjust its style. It learned to be careful and safe during bad economic times (like the high-inflation 1970s) and more aggressive when the economy was booming (like the 1990s).

Step 3: Changing the Terrain (Multi-Asset Allocation)

Next, we realized that only investing in the US stock market is like driving a sports car on a muddy mountain road—sometimes you get stuck. We upgraded the system by giving the computer a garage full of different vehicles:

US Stocks (SPY)
Tech Stocks (QQQ)
Gold (GLD)
Bonds (TLT)
Real Estate (VNQ)

We proved that when the stock market enters a major crash, the bot should pivot capital into hard assets or safety nets to keep compounding value.

Step 4: The Blind Talent Scout (Momentum Ranking)

To make this fully realistic, we stripped away the computer’s ability to “look ahead” into the future. We built a system that looks back at the past 12 months to see which assets are moving with the fastest upward velocity (momentum).

The Rule: Every year, it checks the scoreboard, automatically kicks out the losers, and splits your money 50/50 into the top two winning assets.

Step 5: Future Updates

While the baseline system is highly capable, the upcoming roadmap includes several key structural upgrades to prepare it for advanced market environments:

Dynamic Position Sizing
Instead of a rigid 50/50 split, the system will allocate more money to the asset with the smoothest trend and less to the highly volatile one.
Alternative Macro Filters
We will test replacing the basic stock market filter (RSI / EMA) with real-world economic indicators, like inflation tracking and interest rate moves, to predict market crashes earlier.
Expanded Asset Pool
To capture even more diverse trends, we plan to add international stocks, emerging markets, and broad commodities (like oil and agriculture) to the selection pool.