One-Period Rating Transition Matrices

Why One-Period Rating Transitions Matter

A credit rating summarizes an assessment of an obligor’s credit quality. Over a specified period, an obligor may retain its rating, move to a stronger or weaker rating, or default. A one-period transition matrix organizes the probabilities of these possible outcomes by starting rating. It lets an analyst answer a precise question: given the obligor’s rating now, what is the probability of each rating state at the end of the period?

The period must be stated, such as one year or one quarter, because transition probabilities depend on the horizon. The matrix also depends on the rating categories and the modeled population or assumptions. It is a probability model, not a guarantee that an individual obligor will follow any particular path.

Definition: For an ordered set of rating states, a one-period transition probability $P_{ij}$ is the probability that an obligor has destination rating $j$ at the end of the period, conditional on having origin rating $i$ at the start: $$ P_{ij}=P(R_{t+1}=j\mid R_t=i). $$ A one-period transition matrix $P$ places these probabilities in a table, with origin ratings in rows and destination ratings in columns. Each row is a probability distribution over the possible end-of-period states.

For example, suppose the modeled states are A, BBB, BB, and D, where D denotes default and A is stronger credit quality than BBB, which is stronger than BB. A row for origin rating BBB gives the probabilities of ending the period in A, BBB, BB, or D, respectively. Ratings must appear in the same order in both the rows and the columns; otherwise, the entries can be misread.

Constructing and Checking the Matrix

The analyst first specifies the horizon and the rating states, including how default is represented. Then each origin-rating row is filled with the probability of each possible destination. Since the destinations are mutually exclusive and collectively exhaustive for an obligor over the period, the entries in every row must sum to one. A row that does not sum to one may contain an omitted state, a transcription error, or probabilities rounded so heavily that the discrepancy needs explanation.

For a one-period model, it is common to treat default as an absorbing state: an obligor that is already in D remains in D at the end of the period. This is a modeling convention for the rating-state table, not an assertion that recovery or later resolution is impossible. If the model tracks recovery as a separate state, the state definition and matrix must say so explicitly.

Worked Example: Constructing a One-Period Matrix

A hypothetical annual rating model uses the states A, BBB, BB, and D, in that order. Its stated one-year probabilities are as follows: from A, the probabilities of remaining A, moving to BBB, moving to BB, and defaulting are 0.82, 0.12, 0.04, and 0.02. From BBB, the probabilities of moving to A, remaining BBB, moving to BB, and defaulting are 0.08, 0.78, 0.10, and 0.04. From BB, the probabilities of moving to A, moving to BBB, remaining BB, and defaulting are 0.02, 0.13, 0.75, and 0.10. Use the absorbing-default convention to complete the matrix. For an obligor currently rated BBB, find the probability of an upgrade, the probability of a downgrade to a nondefault rating, the probability of default, and the probability of any rating change.

Method. Place each set of probabilities in the row corresponding to its starting rating, keeping the destination columns in the fixed order A, BBB, BB, D. The D row is $(0,0,0,1)$ under the absorbing-default convention.

P= \begin{pmatrix} 0.82&0.12&0.04&0.02\\ 0.08&0.78&0.10&0.04\\ 0.02&0.13&0.75&0.10\\ 0&0&0&1 \end{pmatrix}

Row check. The three supplied rows sum to $0.82+0.12+0.04+0.02=1.00$, $0.08+0.78+0.10+0.04=1.00$, and $0.02+0.13+0.75+0.10=1.00$. The default row also sums to $1.00$.

For an origin rating of BBB, the relevant row is $(0.08,0.78,0.10,0.04)$. A is the only stronger nondefault rating, so the upgrade probability is $0.08$. The downgrade to a nondefault rating is the probability of moving to BB, $0.10$. The default probability is $0.04$. Any rating change includes an upgrade, a nondefault downgrade, or default, so $$ 0.08+0.10+0.04=0.22. $$ Equivalently, subtract the unchanged-rating probability from one: $1-0.78=0.22$. The results are consistent. Common mistake avoided: counting default as a nondefault downgrade when the question asks specifically for a downgrade to a rated state.

Reading a Row as Conditional Probabilities

For an obligor whose starting rating is known, use that rating’s row directly. The diagonal entry $P_{ii}$ is the probability of retaining the same rating. The default column gives the one-period default probability conditional on the origin rating. The entries in better-rated columns give the upgrade probability; entries in weaker-rated, nondefault columns give the probability of a nondefault downgrade.

The term “downgrade” needs care. Some questions use it to mean movement to any worse state, including default; others distinguish a downgrade to a lower nondefault rating from default. State which interpretation is being used. A reliable approach is to report the nondefault downgrade and default probabilities separately, then add them only if the question asks for deterioration including default.

The probability of any rating change is one minus the diagonal entry. This includes both upgrades and all forms of deterioration, including default. It is not the same as the downgrade probability unless upgrades are impossible.

Worked Example: Separating Migration From Default

Use the BBB row of the hypothetical annual matrix above. Calculate (1) the probability of a nondefault outcome at BBB or better, (2) the probability of a downgrade or default, and (3) the probability of a move to BB conditional on the obligor not defaulting.

Method. Read unconditional event probabilities from the BBB row, and use the definition of conditional probability for part (3). The BBB row is $(0.08,0.78,0.10,0.04)$, corresponding to A, BBB, BB, and D.

(1) An outcome at BBB or better means A or BBB, with no default. Therefore, $$ P(\text{A or BBB}\mid \text{origin BBB})=0.08+0.78=0.86. $$

(2) A downgrade or default means BB or D. Therefore, $$ P(\text{BB or D}\mid \text{origin BBB})=0.10+0.04=0.14. $$ As a check, the complementary outcomes are A or BBB, with probability $0.86$, and $0.86+0.14=1.00$.

(3) The conditional probability of moving to BB given no default is the BB probability divided by the probability of no default. The no-default probability is $1-0.04=0.96$. Thus, $$ P(\text{BB}\mid \text{origin BBB, no default})=\frac{0.10}{0.96}=0.104166\ldots $$ To four decimal places, this is $0.1042$. Common mistake avoided: reporting 0.10 for the conditional probability in part (3). The entry 0.10 is unconditional on the end-of-period state; the specified survival condition requires dividing by 0.96.

Why the Row Must Sum to One

The row-sum rule follows directly from the probability structure, not from a special property of credit ratings. Fix an origin rating $i$, and let the possible destination states be $1,\ldots,n$. At the end of the period, exactly one destination state occurs. The events “destination is 1,” through “destination is $n$,” are mutually exclusive and together cover every possible outcome. Conditional on origin state $i$, their probabilities therefore add to one:

\sum_{j=1}^{n}P_{ij} = \sum_{j=1}^{n}P(R_{t+1}=j\mid R_t=i) =1.

The same partition argument justifies common calculations. If an event consists of several destination ratings, add the corresponding entries in the row. If an event is the complement of retaining the origin rating, subtract the diagonal entry from one:

P(\text{rating changes}\mid R_t=i)=1-P_{ii}.

These calculations presume that the listed states are exhaustive for the model. If a state such as withdrawal, restructuring, or an unobserved rating is possible but omitted, the model must explain how it is handled before the row can be interpreted as a complete probability distribution.

Applying the Matrix to a Starting Rating Mix

A matrix row gives probabilities for one obligor conditional on its starting rating. For a group with known numbers of obligors in each starting rating, multiply each row’s probabilities by the number of obligors in that rating and add across starting ratings. This produces expected destination counts. The result need not be an integer: it is an expectation, not a claim about the realized number of migrations.

This calculation uses linearity of expectation. It does not require the default outcomes of different obligors to be independent to calculate the expected count, provided each obligor’s stated transition probabilities apply to its starting rating. Independence or dependence would matter for the distribution of the realized group total, but that is a different question.

Worked Example: Expected Transitions for a Rated Group

A hypothetical portfolio has 50 obligors rated A, 100 rated BBB, and 50 rated BB at the start of the year. Apply the annual transition matrix from the first example. Calculate the expected number of defaults and the expected number of obligors in each destination state.

Method. For each destination, multiply each origin-rating count by the corresponding transition probability and sum. The expected default count is $$ 50(0.02)+100(0.04)+50(0.10)=1+4+5=10. $$ The expected default proportion is $10/200=0.05$, or 5%.

For destination A, the expected count is $$ 50(0.82)+100(0.08)+50(0.02)=41+8+1=50. $$ For destination BBB, it is $$ 50(0.12)+100(0.78)+50(0.13)=6+78+6.5=90.5. $$ For destination BB, it is $$ 50(0.04)+100(0.10)+50(0.75)=2+10+37.5=49.5. $$ The expected number in D is 10, as calculated above.

Check. The expected destination counts sum to $50+90.5+49.5+10=200$, which matches the starting portfolio size. The fractional counts are valid expected values. Common mistake avoided: applying only one starting-rating row to the whole portfolio, rather than weighting each row by the number of obligors that begins in that rating.

Exam Traps and Interpretation

One of the most common errors is confusing the probability of a particular transition with the probability of any migration. For a given origin rating, the diagonal is the probability of no rating change; one minus that entry is the probability of any change. The default-column entry is only the probability of default, not the probability of all downgrades. For example, from BBB in the stated matrix, the default probability is 0.04, the nondefault downgrade probability is 0.10, and the probability of any change is 0.22.

A second trap is using a conditional probability without changing the denominator. A matrix entry is conditional on the starting rating, but it is not automatically conditional on survival, migration, or another end-of-period event. If a question adds such a condition, divide the probability of the joint event by the probability of the conditioning event.

A third trap is reading the wrong row or column. Rows describe starting states; columns describe ending states. Before calculating, name both states in words. For example, “BBB to BB” means the BBB row and BB column, not the BB row and BBB column.

The one-period matrix is defined for a specified horizon and set of states. It describes transitions over that one period. Using it to calculate probabilities over multiple periods requires additional modeling steps and assumptions; a one-period entry alone is not a multi-period probability.

Key takeaway: Fix the horizon and state order, then read each probability from the row for the starting rating and the column for the ending rating. Check that every row sums to one, distinguish default from nondefault downgrade when needed, and apply extra conditioning explicitly.

Check Your Understanding

For each problem, use the stated rating order and interpret probabilities as conditional on the starting rating unless another condition is stated.

A one-period matrix has states A, BBB, BB, D. Its BBB row is $(0.06,0.81,0.09,0.04)$. Find the upgrade probability, the nondefault downgrade probability, and the probability of any rating change. (Answer: 0.06, 0.09, and 0.19.)
An obligor starts at BB. Its row is $(0.01,0.12,0.77,0.10)$ in the order A, BBB, BB, D. Find the probability of remaining BB or improving, and the probability of default. (Answer: 0.90 and 0.10.)
For the BBB row $(0.06,0.81,0.09,0.04)$, find the probability of ending in BB conditional on not defaulting. Give the answer to four decimal places. (Answer: $0.09/0.96=0.0938$.)
A hypothetical group has 40 A-rated obligors and 60 BBB-rated obligors. Their one-period default probabilities are 0.015 and 0.035, respectively. Calculate the expected number of defaults and the expected default proportion. (Answer: $40(0.015)+60(0.035)=2.7$ expected defaults; $2.7/100=0.027$, or 2.7%.)
A proposed row for origin rating BB has probabilities 0.02 to A, 0.10 to BBB, 0.78 to BB, and 0.06 to D. Identify the row check result and, if the missing probability is assigned to D, state the corrected default probability. (Answer: the entries sum to 0.96, so the row is short by 0.04; assigning the missing probability to D gives a corrected default probability of 0.10.)

The matrix approach is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus. Its central discipline is straightforward but essential: identify the origin state, identify the destination event, and use the matching conditional probability without confusing a one-period transition with a differently conditioned or longer-horizon quantity.

What You'll Learn