Nash Equilibrium explained for betting

In one sentence

A Nash equilibrium is a set of strategies, one per player, where nobody can do better by changing their own strategy while the others keep theirs.

How it works

Many sporting moments are games in the strict sense: a penalty taker and a goalkeeper, a server and a returner, a bowler and a batter. If one side becomes predictable, the other exploits it. The stable outcome is often a mixed strategy, where each side randomises in just the right proportions to make the opponent indifferent.

The idea, formalised by John Nash in 1950, lets you calculate what a well-played contest should look like, including the probability of each outcome. For a bettor, that gives a fair price for events such as "penalty scored" without needing years of data on each player.

It also applies to markets. Traders, bots and market makers are playing a game with each other, and thinking about what the other side will do in response to your strategy is often more useful than a model that ignores them.

The maths

For a two-choice game where the kicker picks left or right and the keeper dives left or right:

q^{*} = \frac{s_{LR} - s_{RR}}{(s_{LR} - s_{RR}) + (s_{RL} - s_{LL})}

s with two letters: the scoring probability for kicker direction then keeper dive, for example s LR means kick left, keeper dives right.
q star: the proportion of the time the keeper should dive left so the kicker gains nothing by favouring either side.
The kicker's mix is found the same way, making the keeper indifferent.

In plain English: each side mixes its choices so that the other side's options all give the same result, leaving nothing to exploit.

Worked betting example

Suppose, for a particular taker, the scoring chances are:

Kicker / keeper	Keeper dives left	Keeper dives right
Kick left	60%	95%
Kick right	90%	70%

Step by step:

Keeper's mix: dive left q = (0.95 − 0.70) ÷ (0.25 + 0.30) = 0.25 ÷ 0.55 ≈ 45.5%.
Kicker's mix: kick left x = (0.90 − 0.70) ÷ (0.20 + 0.35) = 0.20 ÷ 0.55 ≈ 36.4%.
Scoring probability: 0.60 × 0.455 + 0.95 × 0.545 ≈ 79.1%, and the same from the other direction.
Fair odds for "penalty scored": 1 ÷ 0.791 ≈ 1.26.

If an in-play market offers 1.40 on the penalty being scored, it implies about 71%. Backing £50 at 1.40 has an expected value of 0.791 × £20 × 0.98 − 0.209 × £50 ≈ £5.05 after 2% commission, if your scoring table is right. That "if" is doing a lot of work, as the pitfalls below explain.

Where it's good

Pricing set-piece events in-play, such as penalties, where the contest is close to a clean two-player game.
Tennis serve direction and similar repeated duels.
Thinking about competitive trading: if every bot uses the same signal, the edge disappears, and equilibrium thinking predicts that.
Market-making strategy, where your quotes must hold up against traders who know more than you.
Understanding why obvious patterns in sport rarely persist once opponents notice them.

Limitations and pitfalls

The inputs are the hard part. The scoring table above is an assumption; real values vary by player, keeper and pressure, and are poorly estimated from small samples.
Real players are not perfect randomisers. Research such as Palacios-Huerta's 2003 study of penalties found play broadly consistent with equilibrium, but individual players can be predictable.
Most sporting situations have many more options and hidden information, making the equilibrium hard or impossible to compute exactly.
Equilibrium describes how rational opponents play, not how they actually play. Exploiting mistakes can beat equilibrium, and assuming equilibrium can miss that.
In-play markets suspend the moment a penalty is given, so you are pricing what happens when the market reopens, not a price you can hold on to.
Markets usually price these events well already; equilibrium maths is rarely an edge on its own.

How to build it

Python: nashpy for two-player games; scipy.optimize.linprog for zero-sum games of any size.
Data: outcome rates for each pair of choices, such as penalty direction data, with enough samples per cell.
Tip: use shrinkage towards league-wide averages for individual players, because single-player samples are tiny.

Market making - a strategic game against better-informed traders.
Informed trader models - game theory applied to who is on the other side of your bet.
Behavioural biases - where real players and punters depart from equilibrium.
Expected value - turning the equilibrium probability into a betting decision.
Reinforcement learning - learning strategies in games too complex to solve by hand.