Uncertainty in RAROC Estimates

Why Uncertainty Matters in RAROC

A RAROC estimate is a ratio, so uncertainty in either its earnings numerator or its allocated-capital denominator can materially affect a decision. A project that appears to clear a required return using central estimates may fail the hurdle when expected losses are higher, expenses are larger, or required capital is greater than estimated. The relevant question is not only “What is the estimated RAROC?” but also “How much does that estimate depend on uncertain inputs?”

This tutorial treats uncertainty using input ranges and sensitivities. These tools make the effect of imperfect estimates visible without implying that a range has a particular probability. They are aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Definition: For a business decision measured on a consistent time horizon and accounting basis, let \(R\) be revenue or other income attributable to the business, \(L\) its expected loss, \(E\) its expenses, and \(K\) its allocated capital. Define net income for the RAROC calculation as \(N=R-L-E\). When \(K>0\), the estimated RAROC is \(r=N/K\). If the chosen convention includes additional income, costs, or tax adjustments, include them consistently in \(N\); do not mix numerator and denominator bases.

All monetary quantities in the examples are in millions of dollars unless stated otherwise. A range such as \(L\in[7.2,8.0]\) means that the estimate is being assessed over that stated interval. It is not, by itself, a confidence interval or a statement that every value is equally likely. If the ranges for separate inputs are combined as if all endpoints could occur together, the resulting range is a conservative rectangular envelope unless the organization has established that those combinations are feasible.

From Input Ranges to a RAROC Range

Suppose revenue, expected loss, and expenses have ranges \(R\in[R_L,R_U]\), \(L\in[L_L,L_U]\), and \(E\in[E_L,E_U]\). With no additional information restricting their combinations, the corresponding net-income bounds are

N_L=R_L-L_U-E_U,\qquad N_U=R_U-L_L-E_L.

If allocated capital is also bounded by \(0<K_L\le K\le K_U\), the minimum and maximum RAROC over the rectangular set of inputs occur at its corners. A general expression that also works when the net-income interval includes negative values is

r_L=\min\left\{\frac{N_L}{K_L},\frac{N_L}{K_U}, \frac{N_U}{K_L},\frac{N_U}{K_U}\right\},\qquad r_U=\max\left\{\frac{N_L}{K_L},\frac{N_L}{K_U}, \frac{N_U}{K_L},\frac{N_U}{K_U}\right\}.

When \(0\le N_L\le N_U\), these expressions simplify to \(r_L=N_L/K_U\) and \(r_U=N_U/K_L\). This is because, for nonnegative income, a smaller numerator or larger denominator lowers the ratio. If net income can be negative, increasing capital makes the ratio less negative, so it is safer to compare all four corners rather than use the positive-income shortcut.

Decision rule: For a positive hurdle rate \(h\), a decision is robustly above the hurdle over the assessed range if \(r_L\ge h\). It is robustly below if \(r_U<h\). If \(r_L<h\le r_U\), the range straddles the hurdle: the decision is sensitive to the uncertain estimates and should not be described as a robust pass.

Worked Example: Testing a Project Against a Hurdle

A hypothetical underwriting initiative is expected to generate revenue of \(18.0\). Expected losses are estimated between \(7.2\) and \(8.0\), expenses between \(2.0\) and \(2.4\), and allocated capital between \(40\) and \(46\). The required RAROC is \(18\%\). Calculate the central estimate and the range, then assess the decision.

Method. First calculate the numerator bounds, then divide by the appropriate capital endpoints. This avoids treating an uncertain ratio as if it were a ratio of independently chosen central values.

For a central estimate, use expected loss \(7.6\), expenses \(2.2\), and capital \(43\). Thus,

N=18.0-7.6-2.2=8.2,\qquad r=\frac{8.2}{43}=0.1906977\ldots

The central RAROC is \(19.07\%\), rounded to two decimal places. For the lower net-income bound, use the lowest revenue and the highest loss and expense estimates:

$$ N_L=18.0-8.0-2.4=7.6. $$

For the upper bound, use the highest revenue and the lowest loss and expense estimates. Revenue is fixed here:

$$ N_U=18.0-7.2-2.0=8.8. $$

Both numerator bounds are positive, so the lowest ratio pairs the lower numerator with the highest capital, while the highest ratio pairs the higher numerator with the lowest capital:

r_L=\frac{7.6}{46}=0.165217\ldots,\qquad r_U=\frac{8.8}{40}=0.22.

The assessed RAROC range is \(16.52\%\) to \(22.00\%\), rounded to two decimal places. It straddles the \(18\%\) hurdle. The central estimate clears the hurdle, but the project is not a robust pass over the stated range. As a check, \(7.6/46\) is below \(18\%\), while \(8.8/40\) is above it.

Why the Endpoint Calculation Works

For a fixed positive capital amount, \(N/K\) increases as \(N\) increases. For a fixed numerator, the effect of increasing \(K\) depends on the sign of \(N\). The partial derivatives make the directions explicit:

\frac{\partial r}{\partial N}=\frac{1}{K}>0,\qquad \frac{\partial r}{\partial K}=-\frac{N}{K^2}.

If \(N>0\), the second derivative expression is negative, so increasing capital lowers RAROC. If \(N<0\), it is positive, so increasing capital makes the negative RAROC less negative. The ratio is monotone in each input when the other is fixed. Consequently, over a rectangular range with \(K>0\), a minimum and maximum can be found among the four combinations of numerator and capital endpoints. This justifies the corner calculation, including the case where net income changes sign.

Hurdle Rates and Maximum Supportable Capital

A hurdle comparison can also be rearranged to show how much capital a given amount of positive net income can support. Starting with \(N/K\ge h\), assume \(K>0\), \(N>0\), and \(h>0\). Multiplying by \(K\) and dividing by \(h\) preserves the inequality and gives

\frac{N}{K}\ge h \quad\Longleftrightarrow\quad N\ge hK \quad\Longleftrightarrow\quad K\le \frac{N}{h}.

Thus, for \(N>0\) and \(h>0\), the largest capital amount consistent with meeting the hurdle is \(K_{\max}=N/h\). The conditions matter. If \(N>0\) and \(h=0\), every positive \(K\) produces positive RAROC and meets the zero hurdle; there is no finite largest capital amount. If \(N\le0\) and \(h>0\), no positive capital amount can produce RAROC at or above that positive hurdle. A zero capital denominator does not define RAROC.

Worked Example: Capital Capacity Under an Uncertain Numerator

A hypothetical service expansion has expected net income estimated between \(6.3\) and \(7.1\). Management requires a \(18\%\) RAROC. Existing allocated capital is \(32\). Determine the maximum capital consistent with the hurdle using the lower income estimate, and assess whether the existing allocation meets the hurdle throughout the income range.

Method. Since the hurdle is positive and both net-income estimates are positive, calculate capital capacity using \(K_{\max}=N/h\). To test whether the current allocation is robust, use the lower net-income estimate.

Using \(N_L=6.3\) and \(h=0.18\),

K_{\max,L}=\frac{6.3}{0.18}=35.0.

The lower-estimate income supports up to \(35.0\) of capital, so an allocation of \(32\) is within that limit. Directly checking the low and high income endpoints gives

r_L=\frac{6.3}{32}=0.196875,\qquad r_U=\frac{7.1}{32}=0.221875.

The range is \(19.69\%\) to \(22.19\%\), rounded to two decimal places, and its lower endpoint exceeds \(18\%\). The allocation meets the hurdle throughout the stated income range. As a check, the lower-end RAROC is equivalent to the capital-capacity test: \(32<35.0\).

Sensitivity to Individual Estimates

A range describes a combined set of possible outcomes. A sensitivity calculation instead asks how much RAROC changes when one input changes while the others are held fixed. This is useful for identifying which assumptions deserve closer review, but it does not by itself measure the probability of a change or the likelihood that several changes occur together.

With \(r=(R-L-E)/K\), the local effects of small input changes are summarized by

\frac{\partial r}{\partial R}=\frac{1}{K},\qquad \frac{\partial r}{\partial L}=-\frac{1}{K},\qquad \frac{\partial r}{\partial E}=-\frac{1}{K},\qquad \frac{\partial r}{\partial K}=-\frac{N}{K^2}.

These derivatives are per unit of the relevant monetary input when all quantities use the same units. For example, if amounts are in millions of dollars, the derivative with respect to loss is the change in the decimal RAROC per additional million dollars of loss. For a material change, calculate the new ratio directly; the derivative is a local approximation, not an exact finite-change formula.

Worked Example: Separating Loss and Capital Sensitivities

A hypothetical portfolio has revenue \(25\), expected loss \(12\), expenses \(3\), and allocated capital \(50\). Calculate its base RAROC, then recalculate it for (a) an additional \(0.5\) of expected loss, (b) an additional \(5\) of capital, with all other inputs unchanged. Use the exact ratios and compare the capital change with the local sensitivity.

Method. Establish the base numerator and ratio, then change one input at a time. This makes clear which assumption caused each change.

The base net income and RAROC are

N=25-12-3=10,\qquad r=\frac{10}{50}=0.20=20.00\%.

With expected loss increased by \(0.5\), net income becomes \(10-0.5=9.5\), and capital remains \(50\):

r_{\text{higher loss}}=\frac{9.5}{50}=0.19=19.00\%.

With capital increased by \(5\), net income remains \(10\) and capital becomes \(55\):

r_{\text{higher capital}}=\frac{10}{55}=0.181818\ldots=18.18\%.

At the base point, the derivative with respect to capital is \(-10/50^2=-0.004\) per million dollars. For a \(5\)-million increase, the first-order change is approximately \(-0.004(5)=-0.020\), or a decrease of \(2.00\) percentage points, giving an approximate RAROC of \(18.00\%\). The exact result is \(18.18\%\); the difference occurs because the derivative approximation is local. The loss sensitivity is exact in this example because the denominator is fixed: \(0.5/50=0.01\), a one-percentage-point decrease. The calculations check: more loss or more capital lowers the positive RAROC.

Common Mistakes and Interpretation

Using only the central estimate. A central RAROC above the hurdle does not establish that the decision is robust. Compare the full assessed range with the hurdle.
Combining the wrong endpoints. When net income is positive, the low ratio uses low net income and high capital, not low net income and low capital. In the first worked example, \(8.8/46=19.13\%\) is not the lower bound; the correct lower bound is \(7.6/46=16.52\%\).
Applying the positive-income shortcut when income may be negative. If the numerator interval crosses zero, evaluate all four endpoint ratios. For negative net income, increasing capital can make the ratio less negative.
Treating an interval as a probability statement. A range of plausible estimates does not imply a confidence level unless a statistical method and its assumptions support that interpretation.
Ignoring measurement consistency. Revenue, losses, expenses, and capital must refer to compatible periods and business scope. Mixing annual income with capital measured on a different basis can make the ratio misleading.
Dividing by a hurdle without checking its sign. The formula \(N/h\) is a finite maximum-capital limit only when \(N>0\) and \(h>0\). With \(N>0\) and \(h=0\), every positive capital amount meets the hurdle, so no finite maximum exists.

A useful conclusion distinguishes three cases. A decision is robustly attractive when the lower RAROC bound meets the hurdle. It is robustly unattractive when even the upper bound misses the hurdle. When the interval crosses the hurdle, the result is fragile with respect to the stated uncertainty: the estimate may still support proceeding, but the decision depends on which values prove appropriate. In that case, identify the inputs that drive the range and determine whether better estimation or a different allocation would change the decision.

Key takeaway: RAROC uncertainty should be assessed through both the numerator and capital denominator. State the basis, calculate a defensible range, and compare its endpoints with the hurdle. Use sensitivities to locate important assumptions, while keeping estimation ranges distinct from probability-based statements.

Check Your Understanding

For each problem, use consistent monetary units and check whether the conclusion follows from the full range, not just a central estimate.

A hypothetical product has revenue \(11\), expected losses of \(4.5\), expenses of \(1.5\), and capital of \(25\). Calculate its RAROC and state whether it meets a \(19\%\) hurdle. (Answer: \(20.00\%\); it meets the hurdle.)
A hypothetical portfolio has fixed revenue \(16\), expected loss between \(5\) and \(6\), expenses between \(1.5\) and \(2.0\), and capital between \(35\) and \(40\). Calculate the conservative RAROC range and assess it against an \(18\%\) hurdle. (Answer: \(20.00\%\) to \(27.14\%\); robustly above the hurdle.)
A project has net income between \(9.6\) and \(11.0\), capital between \(60\) and \(64\), and a \(16\%\) hurdle. Calculate the lower and upper RAROC bounds and classify the result. (Answer: \(15.00\%\) to \(18.33\%\); the range straddles the hurdle, so it is fragile.)
For a hypothetical business, revenue is \(20\), expected loss is between \(8\) and \(9\), expenses are between \(2\) and \(2.5\), and capital is between \(40\) and \(50\). Calculate the RAROC range and determine whether it robustly meets an \(18\%\) hurdle. (Answer: \(17.00\%\) to \(25.00\%\); it does not robustly meet the hurdle because the range straddles \(18\%\).)
A business has positive net income estimated at \(6.3\) and a positive hurdle of \(18\%\). Find its maximum supportable capital. Then state whether a current allocation of \(36\) meets the hurdle. (Answer: \(6.3/0.18=35.0\); the allocation of \(36\) does not meet the hurdle.)

What You'll Learn