ASA common mistakes
ASA · Understand the traps and how to avoid them.
On Probability (P) problems involving a mixture or a piecewise density, the candidate writes one integral over the full support and evaluates it with a single density expression, rather than splitting at the change point or conditioning on the mixture component.
Practice with single-form densities such as exponential, uniform, and normal creates a pattern-match; under time pressure the candidate reads the support and ignores the piecewise definition or the mixing weight.
What to do instead
Before calculating, redraw the support as a number line, mark every breakpoint and mixture component, and write separate integrals or a law-of-total-probability sum. Check that the total probability equals one as a quick audit.
In annuity and loan amortization problems, the candidate substitutes an annual effective interest rate directly into a monthly or quarterly formula and uses the number of years as the number of payment periods.
Textbook formulas are often memorized with i and n without tracking the compounding or payment unit; exam wording gives an annual rate and a term in years.
What to do instead
Write a unit check at the top: payment period, rate period, and number of periods must match. Convert i to the payment period using (1+i)^(1/m)-1 for effective rates or i^(m)/m for nominal rates, and multiply years by m. Then verify the first payment's interest accrual.
When a problem specifies a fractional-age assumption such as UDD, constant force, or Woolhouse, or specifies select mortality, the candidate ignores it and uses the standard ultimate life table directly.
The candidate focuses on the table values and treats the assumption sentence as boilerplate; under time pressure, they default to the first formula they recall.
What to do instead
Before touching the table, copy the stated assumption and the select period into the solution. Choose the matching formula and, for select models, use the select-and-ultimate table with the correct duration since issue.
In reserve and cash-flow problems, the candidate calculates a gross premium reserve by subtracting all expenses at time 0 and then discounting the remaining net cash flows, instead of placing each expense at its actual payment time and matching assets and liabilities.
Net premium reserve formulas are simpler and get overlearned; expense timing and per-policy initial costs are easy to lump together.
What to do instead
Lay out a timeline with columns for time, premium, expense, benefit, and survival probability; put every cash flow at its actual time, then take the expected present value. For gross reserves, verify the equivalence principle at issue and check the recursion from one duration to the next.
In regression and GLM output, the candidate interprets a 95% confidence interval for the mean response as a 95% prediction interval for a new observation, or treats a p-value as the probability the null hypothesis is true.
Both intervals look similar in software output; p-value language in news and prior courses reinforces the inverse-probability misinterpretation.
What to do instead
State the estimand before reading the output: mean response versus new observation. Use the wider interval with the additional variance term for prediction. For p-values, write 'probability of data this extreme if H0 true' and avoid inverse statements; check the null and alternative before concluding.
The candidate selects a model by comparing training-set error or R-squared across models, then reports that training performance as if it were out-of-sample performance.
Software reports fit statistics by default; candidates want to show a strong model and skip the holdout or cross-validation step under time pressure.
What to do instead
Split data before any preprocessing, use a holdout or k-fold cross-validation, and choose models only on validation error. Refit the chosen model on training data, then report test error once; never tune on the test set.
In a predictive analytics assessment, the candidate treats a model's feature importance or coefficient sign as proof of a causal effect and recommends an intervention without considering confounding.
Feature importance is easy to produce and visually persuasive; assessment prompts ask for business recommendations, which invites causal language.
What to do instead
Separate prediction from causal inference: state that importance is associational, list plausible confounders, and recommend an experiment or quasi-experimental design before any intervention. If the task is prediction only, frame recommendations around scoring and monitoring.
The candidate applies the central limit theorem to a sample of size 8 from a heavily skewed distribution, or uses a normal approximation for a binomial with np<5, without checking conditions.
The CLT is taught as a universal tool and software makes normal calculations easy; candidates memorize n>=30 as a magic number but ignore skewness and dependence.
What to do instead
Check the sampling distribution: for small n, use the exact distribution or a transformation; for binomial, verify np and n(1-p) are adequate or use exact methods. State the condition and the approximation error direction before computing.
The candidate answers a supply-and-demand question by shifting the demand curve when the event changes a non-price determinant of supply, or by moving along a curve when the event is a price change.
The graph looks similar and the labels are easy to swap; exam wording sometimes embeds the determinant in a sentence.
What to do instead
Classify the shock first: price change means movement along; a non-price determinant such as income, input costs, technology, expectations, or number of buyers or sellers means shift. Write the determinant and the affected curve before drawing; then trace the new equilibrium price and quantity.
The candidate classifies a cash flow as operating when it is investing or financing, or compares accounting book value with market value in a valuation problem.
Introductory accounting emphasizes net income; candidates memorize the statement categories but not the decision rules, and finance formulas sometimes use book values in ratios.
What to do instead
Use the cash-flow classification test: operating = core business transactions; investing = long-term assets; financing = capital providers. For valuation, convert to market values or cash flows and state which basis is used; reconcile net income to operating cash flow with non-cash items and working-capital changes.