Classic Strategies Under The Microscope

Flat Betting, Martingale, and Reverse Martingale: Do They Ever Work?

This post is part of a multi-part series on building a roulette simulator and using AI to test betting strategies.

• Start from the beginning: The House Always Wins? I Taught 3 AI Agents to Beat Roulette Anyway
• Explore the code: GitHub – rossautomatedsolutions/roulette-simulator

---

Before jumping into AI-driven betting strategies, I started with a baseline: the most common betting systems players use in real casinos — Flat Betting, Martingale, and Reverse Martingale.

These are often passed down as “systems” or “strategies” despite being mathematically doomed. I wanted to analyze them in a rigorous, simulation-based way.

---

Simulation Setup

I simulated American-style roulette (38 slots: 0, 00, and 1–36) under the following rules:

• Starting bankroll: $1,000
• Min bet: $20
• Max bet (where applicable): $500
• Bets were placed on red (outside bet) unless otherwise defined
• Simulation ends after 1,000 spins or when bankroll hits $0
• Each strategy was run 100 times with randomized spin outcomes

---

Strategies I Tested

Flat Betting

The most common “low risk” strategy. Bet a fixed amount each spin, regardless of wins or losses. In this case, I always bet $20 on red.

Why some people use it:
It avoids the exponential risk of Martingale while keeping betting consistent.

Why it fails:
The house edge guarantees a slow bleed over time.

Expected Value Proof:

• P(win) = 18/38
• P(loss) = 20/38
• Payout = 1:1
• EV per $1 bet = (18/38 × $1) + (20/38 × -$1) = -$0.0526
• So for a $20 bet: EV ≈ – $1.05 per spin

That negative expected value builds slowly, but inevitably.

---

Martingale

Double your bet after every loss. Reset to base bet after each win.

Why some people use it:
It “guarantees” profit eventually, since one win recovers all previous losses.

Why it fails:
Losing streaks cause bet sizes to grow exponentially. A few losses in a row wipes out your bankroll or hits the table max.

---

Reverse Martingale (Paroli)

Double your bet after each win. Reset to base after a loss.

Why some people use it:
It limits exposure during cold streaks and presses the advantage during hot ones.

Why it fails:
Streaks are rare and unpredictable. One loss wipes out the entire series of wins.

---

Simulation Results (Averaged over 100 Runs)

Strategy	Avg Spins	Avg Final Bankroll	% Profitable Runs	% Bankruptcies	Max Drawdown (avg)
Flat Bet	775	$204	5%	71%	$1,058
Martingale (Limited)	210	$151	4%	96%	$1,897
Martingale (Unlimited)	199	$310	4%	96%	$2,382
Reverse Martingale (Limited)	260	$267	7%	91%	$1,827
Reverse Martingale (Unlimited)	96	$0	0%	100%	$11,045

---

Observations

• None of the strategies produced consistent profits. The average P&L for all of them was negative.
• Flat Betting had the highest survival time but still lost money ~95% of the time.
• Martingale strategies occasionally posted large gains, but 96% of runs ended in collapse.
• Reverse Martingale (Unlimited) had a 100% bankruptcy rate.
• The illusion with Martingale is that small wins appear often — but one bad streak ends the game.

---

Takeaway

These betting systems are built on flawed assumptions about streaks and variance. They change how you lose — slowly or suddenly — but not whether you lose.

They served as a useful control group. The real question was whether a learning algorithm could outperform these static systems by reacting to history and adapting bet selection.

That’s what I tackled next — with Q-learning.

---

Continue to Post 3: Teaching AI to Bet – The Q-Learning Roulette Agent