Attempts at Building a Trading Algo
Early in middle school, my dad created a paper-trading account for me on Wall Street Survivor to learn the basics of trading. At first, I didn't know what to do: I would randomly choose ticker symbols that I liked, putting my fake money into tickers like BOOM and CAKE. Eventually, after learning from mistakes, reading, and attending seminars, I eventually reverse-engineered my way into a basic understanding of the stock market.
Note: last I checked, the paper trading account had 10x with the NVDA and MU I bought in middle school...
As I've had less time to follow the market, I've been increasingly interested in automating screening and trading approach I use. In 2023, I started trying to build an algorithm based on the technical indicators I've historically leveraged manually, and the results are below.
V1 & V2
I spent a lot of time learning how to build any sort of algorithm for trading in Python to make V1 and V2. I landed on Alpaca.py as a platform that I could pull live data from and place trades through for free.
V1 was built around a pretty simple idea: use technical indicators to predict stock price movement one hour ahead, then automatically make buy or sell decisions based on that prediction. I trained a Random Forest model using indicators like RSI, MACD, moving averages, stochastic oscillators, and momentum-based features as inputs.
While the results looked promising in backtesting, once I started paper trading the strategy live, I realized that seemingly minor issues were erasing most of the theoretical gains. Small delays in price updates, spreads between bid and ask prices, and market order execution all created slippage and losses.
V2 was an attempt to fix those problems by rewriting the live price update and order execution logic to reduce slippage. While performance improved, the core issue of too much noise and slippage amidst minor price movements persisted. Looking back, I think this architecture could have worked if it was trained on "abnormal" market moves instead of trying to predict and trade every time period.
V3
After spending too much time fighting noise, I decided to completely scrap the active trading logic and predictive capabilities. I shifted back to building a purely screening and signal-generation system based on the technical setups I already used manually. V3 focuses on identifying stocks that appear attractive on a multi-day swing trading timeframe.
The system now continuously scans stocks for combinations of indicators and trend conditions that historically aligned with the kinds of trades I would personally take, surfacing potential buy/sell opportunities for review. V3 integrates directly into how I actually trade.
<-- Snapshot of signals from V3, scanning the stocks in the Nasdaq 100 on June 5, 2026. Seeing a correction primarily across semiconductor stocks, the trade action taken was to purchase SOXL
Live Testing Log
June 5, 2026 - June 26, 2026
SOXL | Length: 10 days | Gain: 30.0%
FER | Length 11 days | Gain: 5.8%
GEHC | Length: 3 days | Gain: 6.2%