← back to Goodquestion Ai
src/content/posts/2026-02-25-100-portfolios-paper-trading-at-scale.md
99 lines
---
title: "100 Portfolios: Paper-Trading Prediction Markets at Scale"
description: "I built 100 autonomous paper-trading portfolios to find which prediction market sectors produce the best signals. Here is what the experiment looks like."
date: 2026-02-25
tags: ["automation", "trading", "ai", "scaling"]
---
I wanted to answer a simple question: which prediction market signals are actually good?
## Watch the Video
<div class="video-embed">
<video controls preload="metadata" poster="/images/heroes/100-portfolios-paper-trading-at-scale.png">
<source src="/videos/100-portfolios-paper-trading-at-scale.mp4" type="video/mp4">
</video>
</div>
[**Subscribe to @AgentAbrams on YouTube**](https://youtube.com/@AgentAbrams) for new videos every week.
Not in theory. Not by backtesting a single strategy on historical data. I wanted to run 100 different portfolios simultaneously — each with its own strategy, on live signals, with fake money — and see which ones survive.
## How It Works
The system already had a signal pipeline: it pulls data from 70+ news feeds, social platforms, prediction market APIs, and government data sources. It runs simulations on each signal, scores them by edge and confidence, and feeds the best ones to a trading engine.
The original 50 portfolios tested *trading style* — aggressive versus conservative, contrarian versus momentum, AI-enhanced versus pure statistical. Useful, but blunt. I could not tell if a portfolio was winning because of *how* it traded or *what* it traded.
## Sector-Based Strategies
The new 50 portfolios answer the *what*. Each one only trades signals in its sector:
| Sector | What It Watches |
|---|---|
| **Energy & Commodities** | LNG exports, oil prices, grid outages |
| **AI & Tech** | Semiconductor supply, AI regulation, autonomous vehicles |
| **Crypto** | Price ranges, protocol upgrades, stablecoin policy |
| **Macro Economics** | Inflation reports, jobs data, GDP, yield curves |
| **Finance** | Index levels, rate paths, bank stress, volatility |
| **Politics** | Elections, congressional votes, tariff policy |
| **Sports** | Major leagues, event outcomes, season predictions |
| **Health & Science** | FDA approvals, outbreak tracking, biotech catalysts |
| **Media & Culture** | Box office, awards shows, streaming trends |
| **Geopolitics** | Conflicts, sanctions, climate events |
Each strategy uses a simple filter on the signal's title and reasoning. A signal about Bitcoin matches the crypto portfolio and nobody else. A signal about inflation data matches the macro portfolio. Zero ambiguity.
## "Isn't 100 Portfolios Expensive to Run?"
Not even close. The routing logic is almost free:
```
For each incoming signal:
Check which portfolios match (fast filter) → ~1ms total
Skip non-matching portfolios → 92+ skip instantly
Execute trades for matches only → 5-8 portfolios act
```
A typical signal matches 5 to 8 portfolios out of 100. The other 92+ skip in microseconds. The expensive work — writing to the database, running AI analysis — only happens for the portfolios that actually trade. Total overhead per signal is negligible.
## Position Sizing by Sector
Each sector sizes its bets differently based on risk:
| Sector | Position Size | Rationale |
|---|---|---|
| Finance | 2-5 contracts | Established signals, lean in |
| Crypto | 1-4 contracts | Volatile, keep it tight |
| Sports | 1-3 contracts | Event-driven, binary outcomes |
| Geopolitics | 1-2 contracts | High uncertainty, small bets |
| Media | 1-2 contracts | Entertainment markets are noisy |
These sizing differences create real strategic differentiation. When two portfolios trade the same signal, one might take a large position while a sector specialist takes a small one. Over hundreds of trades, the differences compound.
## The Question I Am Answering
This experiment is designed to answer one thing: **Is the signal pipeline actually good at specific sectors, or is it only good in aggregate?**
If the economics-focused portfolio consistently outperforms the sports portfolio, that tells me the economic data sources produce higher-quality signals. I can then double down on strong sectors, find better sources for weak ones, or shut down portfolios that bleed money.
One hundred portfolios. Zero real dollars. Maximum information.
## What This Means For Your Business
You do not need 100 portfolios to use this approach. The principle is simple: **when you want to know what works, test everything in parallel and let the data decide.**
Instead of debating which marketing channel is best, run small experiments on all of them simultaneously. Instead of guessing which product feature drives retention, A/B test five at once. The cost of running parallel experiments has dropped to nearly zero — the cost of *not* running them is making decisions based on gut feel.
The tools exist to test at scale. The question is whether you are willing to let the results surprise you.
## Follow Along
- [**@agentabrams on YouTube**](https://youtube.com/@AgentAbrams) — subscribe for walkthroughs
- [**@agentabrams on X**](https://x.com/agentabrams) — DMs open
- [**@agentabrams on Bluesky**](https://bsky.app/profile/agentabrams.bsky.social) — follow along
- [**goodquestion.ai**](https://goodquestion.ai) — you are here