A base rate is how often an event occurs in a specified group before you use the particular evidence being assessed. It supplies context for interpreting an alert or a striking story. An alert that catches most events can still be wrong more often than right when the event is rare.
A fraction divides a part by its whole: 20 events among 1,000 cases is 20 ÷ 1,000 = 0.02, or 2%. A percentage means a count per hundred. The group used for comparison is the reference group. For investment research, a one-year default rate among a specified set of bond issuers answers a different question from a ten-year default rate or a rate among already distressed issuers. Default means failing to meet a specified debt obligation.
A historical frequency is an estimate from a sample, not a known law of nature. State how issuers were selected, how default was defined, which dates were covered and how missing outcomes were treated. A broad rate may be a poor starting point for an issuer with materially different characteristics. A carefully chosen reference group is part of the reasoning, not a footnote.
Conditional probability means the chance within a restricted group. “Alert given default” counts alerts among cases that default. “Default given alert” counts defaults among cases that trigger an alert. Those groups have different denominators. Confusing the two, or ignoring the starting frequency when interpreting evidence, is the base-rate fallacy.
Illustrative one-year example: assume 1,000 comparable issuer-years, each one issuer observed for one year. Exactly 20 end in default and 980 do not. An alert issued at the start of each year flags 80% of the default cases and 10% of the non-default cases. It flags 16 default cases and 98 non-default cases: 114 alerts altogether. Among alerts, the default fraction is 16 ÷ 114 ≈ 14.04%, not 80%. These are assumed teaching counts, not a measured financial dataset.
The alert raises the chance from the assumed 2% starting rate to about 14.04%. It is informative under these assumptions even though most alerts do not precede default. Both statements can be true. A low starting frequency leaves a large non-default group from which false alerts can arise.
A prior is a probability before adding the evidence currently under discussion. A relevant base rate can help set that prior; priors can also reflect other justified knowledge. A posterior is the probability after incorporating the evidence. Bayes’ rule connects the two using likelihoods: how probable the observed evidence would be in each possible case. The Academy lesson works through that calculation.
A prior that already uses an alert must not be updated again as if that same alert were new information. Changing market conditions can also make old frequencies and alert rates unsuitable for a new group. An exact division cannot repair those inputs. Event probability alone says nothing about the size of a loss, price paid, recovery after default or the investor’s financial circumstances.
No. It is a starting frequency for a specified reference group. Relevant new evidence can change the probability, while a poorly matched group can make the starting rate unsuitable.
No. A detection rate counts alerts among event cases. The fraction of alerts followed by the event also depends on the base rate and the false-positive rate among non-event cases.
No. A single update uses conditional probabilities. Repeated updates require likelihoods that account for evidence already observed; multiplying rates for related alerts as if they were independent can double count information.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.