liveBook · Manning

Maximin method

As explained in Statistics Every Programmer Needs

The maximin method selects the alternative whose worst possible payoff is better than the worst possible payoff of every other alternative.

The maximin method chooses the alternative with the highest worst-case payoff. It suits risk-averse decisions when outcome probabilities are unknown or unreliable. section 10.0 section 10.1

Apply the method as follows:

  • List each alternative's payoff across every possible state of nature. section 10.1.2
  • Find the minimum payoff in each alternative's row. section 10.1.2
  • Select the alternative whose minimum payoff is largest: maximin = max(minimum payoffs). section 10.1.2

Investment choice

For $1 million invested under strong or weak economic conditions, the savings account has the highest minimum payoff: $35,000, compared with −$120,000 for the stock market, −$20,000 for a soccer club, and −$70,000 for a 50/50 combination. Maximin therefore selects the savings account. section 10.1.2

>>> maximin = payoff_table.copy()     #1
>>> maximin['Maximin'] = \
>>>    payoff_table[['Strong Economy', 
>>>                  'Weak Economy']].min(axis = 1)     #2
>>> print(maximin)     #3
             Investment Alternatives  Strong Economy  Weak Economy  Maximin
0                       Stock Market          100000       -120000  -120000
1           Professional Soccer Club           60000        -20000   -20000
2 Stock Market / Soccer Club (50/50)           80000        -70000   -70000
3                    Savings Account           35000         35000    35000

>>> highest_payoff = maximin['Maximin'].max()     #4
>>> print("Highest Payoff:", highest_payoff)     #5
Highest Payoff: 35000

The Python listing copies the payoff table, computes row minima with min(axis=1), prints the results, and extracts the largest minimum payoff. section 10.1.2

How Statistics Every Programmer Needs covers it

How deep it goes
Explains the worst-case rule, gives an implementation procedure, and stops short of optimizing long-term growth.
Examples
Uses an investment payoff table comparing stocks, a soccer club, a 50/50 portfolio, and a savings account.
What it assumes
Assumes a payoff table and possible states of nature, while treating outcome probabilities as unavailable or unreliable.
What you can do after
Apply it via Python by computing row minima and selecting the alternative with the largest minimum payoff.
How it connects
Contrasts with Maximax and Minimax Regret and belongs alongside Expected Value and decision trees.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Statistics Every Programmer Needs. The section links point to the parts of the book it draws on.