What it means
A probability mass function gives probability at each allowed value, and NIST's statistical guidance states that these probabilities must be nonnegative and sum to one. The outcomes are distinct: a portfolio can have zero, one or two defaults, rather than one and a half default events, even though the underlying exposure may have continuous values.
A list of numbers that violates those conditions is not a valid complete distribution. Discrete does not necessarily mean only a few outcomes, since a count can have a large or theoretically unbounded range.
The defining feature is separate possible values rather than a continuous interval. The model must include the relevant outcomes, because leaving out a possible result can understate risk or make the listed probabilities fail to total one, so a scenario set should explain how it represents the full uncertainty.
A binomial model can describe the count of successes across a fixed number of trials under specified assumptions, but the trials need an appropriate probability structure, and applying it to dependent events without adjustment can misstate uncertainty. A Poisson model is another familiar count distribution, used for certain event-occurrence settings with its own assumptions.
The fact that a variable is a count does not automatically prove that a Poisson model is suitable, and the probabilities need evidence or stated assumptions, since historical frequencies can inform them but conditions may change. Expected value summarises the probability-weighted average outcome, and it does not predict the exact next observation, so a model with an expected 1.4 events can still generate only whole event counts in a particular period.
Spread matters alongside the mean, because two distributions can have the same expected count and different chances of unusually high outcomes. Capacity and cash-reserve decisions can depend on those tails.
Counts and financial severity are separate questions, since ten small claims can cost less than one severe claim and a count distribution may need a separate model for the amount associated with each event. Dependence can affect the result as well: defaults or claims can cluster because of a shared shock, and treating events as independent can understate the chance of a large combined outcome.
The time period and exposure should be consistent, because a weekly event count should not be compared with an annual count as if their scales match and changes in the number of customers or loans also affect interpretation. A continuous approximation can help calculation, but it does not change the actual count's possible outcomes, and its accuracy should be checked, especially for small counts or tail probabilities.
For a non-finance manager, identify what is being counted and over which period, then review the probability assumptions, high-count scenarios and the cost per event before using the model. A discrete distribution makes uncertainty explicit, but its usefulness depends on matching the actual business process.
In practice
Real-world examples.
Example
A lender models the number of defaults in a small portfolio. It checks dependence and the size of each exposure rather than translating an expected default count directly into a complete loss forecast. The risk team also tests what happens if two defaults arrive together.
Example
An insurer forecasts the number of claims next month. It keeps a separate severity estimate because the count alone does not show how much cash the claims will require. Reserves are set using both models together.
Example
A service team uses probabilities for zero, one, two or more daily outages. The manager verifies that the categories cover all outcomes and their probabilities sum to one before planning staffing. The final category, two or more, ensures no outcome is left out.
Formula
Calculation
Expected count = sum of outcome x its probability. If zero events has probability 0.50, one has 0.30 and two has 0.20, the probabilities sum to 0.50 + 0.30 + 0.20 = 1, and the expected count is 0 x 0.50 + 1 x 0.30 + 2 x 0.20 = 0.70. An individual observation still has zero, one or two events, not 0.70 of an event.
For spread, the average of the squared outcomes is 0 x 0.50 + 1 x 0.30 + 4 x 0.20 = 1.10, so the variance is 1.10 - 0.70 x 0.70 = 1.10 - 0.49 = 0.61. If each event costs an average of $50,000, the expected loss is 0.70 x $50,000 = $35,000, but the actual loss in a period will be $0, $50,000 or $100,000.Case study
Seen in the real world.
Fictional case: A manager at Linden Support Services, an invented company, budgets for customer incidents using the average count alone. An analyst builds a discrete distribution and finds that the same average is compatible with occasional high-count days. The team then checks whether incidents cluster and combines the count scenarios with response costs.
Staffing and cash planning use the range of outcomes rather than treating the expected count as a guaranteed daily workload. The analyst also records the period and the customer base behind the probabilities, so the figures can be refreshed when the customer base grows. The manager now reviews the probability of a high-count day each quarter, and the board sees both the average and the tail.
Watch out
Common mistakes.
- Using probabilities that are negative, exceed one or do not cover the full outcome set.
- Treating an expected count as the exact next outcome.
- Assuming every count follows the same model while ignoring dependence, exposure and event severity.
Questions
People also ask.
Must the possible values be integers?
Counts usually are, but a discrete variable can use any separate countable values.
Can the expected count be fractional?
Yes. It is a weighted average, not necessarily an outcome that can occur.
Does a count distribution fully measure financial loss?
No. The size and dependence of the losses may require additional modelling.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%