Assessing the Stability of Project Rankings

Why Ranking Stability Matters

A project with the highest estimated net present value is the preferred alternative under the assumptions used in that estimate. That does not by itself show that its lead is dependable. If a modest change in sales, cost, or another important assumption reverses the order, decision-makers should distinguish a narrow ranking from a robust one.

The key quantity is the NPV gap between two alternatives. Assessing ranking stability means examining how that gap changes over plausible assumption ranges or coherent scenarios. It is not enough to note that one project has the larger base-case NPV, or to compare the projects’ uncertainty separately without considering which assumptions they share.

Definition: For two alternatives A and B evaluated on a consistent basis, define the NPV difference as $D=NPV_A-NPV_B$. A has the higher NPV when $D>0$; B has the higher NPV when $D<0$; they tie when $D=0$. A ranking is robust over a stated set of assumptions if the sign of $D$ is the same for every assumption combination in that set.

The usual NPV model discounts the project’s cash flows at the specified rate. For project $j$, with cash flow $CF_{j,t}$ at time $t$ and discount rate $r$, the expression is:

NPV_j=\sum_{t=0}^{n}\frac{CF_{j,t}}{(1+r)^t}.

This comparison presumes the alternatives are evaluated using a consistent discount-rate convention, the same decision horizon, and cash flows defined on the same basis. The analysis below concerns how assumptions affect their relative NPVs; it does not replace checking whether the alternatives are feasible or whether the underlying cash-flow estimates are appropriate.

Compare the Gap With Changes in Value

A base-case gap is a starting point, not a stability conclusion. Sensitivity analysis asks how NPV responds when an input changes. For a ranking decision, the most direct quantity to examine is the corresponding response of $D$. When the estimated effects are linear over the range being tested, the gap can be written as a base-case gap plus the effects of deviations in the assumptions:

D(\mathbf{x})=D(\mathbf{x}_0)+\sum_{k=1}^{m}b_k(x_k-x_{k,0}),

Here $\mathbf{x}_0$ denotes the base-case assumptions and $b_k$ is the change in the NPV difference per unit change in assumption $x_k$, with other assumptions held fixed. This is an exact expression only when the relationship is linear over the range considered; otherwise it is a local approximation or must be replaced by recalculated NPVs. The sensitivity analysis method and any approximation should match the project model.

For independent interval checks, calculate the lowest and highest possible values of $D$ over the stated ranges. If the lowest value is positive, A ranks above B throughout those ranges. If the highest value is negative, B ranks above A throughout. If the interval includes zero, the tested ranges permit a tie or reversal, so the ranking is not established as robust. This test is only as good as the ranges and relationships specified.

Key result: To establish that A remains ahead of B across a specified assumption set, show that the minimum value of $NPV_A-NPV_B$ over that set is greater than zero. A positive base-case gap alone does not establish this.

Worked Example: Testing a One-Input Ranking Range

Two hypothetical projects have base-case NPVs of $180,000 for A and $150,000 for B. Over the range considered, each additional unit of a demand index raises A’s NPV by $4,000 and B’s by $2,000. The base demand index is 100, and the plausible range is 85 to 110. Determine whether A’s ranking is robust.

Method. Calculate the base gap, then model how the gap changes with demand. All NPVs in the calculation are in thousands of dollars. The per-unit effect on the gap is $4-2=2$ thousand dollars.

At the base value, $D(100)=180-150=30$. Thus:

$$ D(Q)=30+2(Q-100). $$

At the low endpoint, $Q=85$, so $D(85)=30+2(85-100)=30-30=0$. At the high endpoint, $Q=110$, so $D(110)=30+2(110-100)=30+20=50$. Because the gap increases with $Q$, its range is from 0 to 50 thousand dollars.

A is never below B on the stated range, but at $Q=85$ the projects tie. Therefore, A’s strict ranking is not robust across the entire range; it is weakly ahead throughout and strictly ahead for $Q>85$. The break-even demand value follows from $0=30+2(Q-100)$, which gives $Q=85$.

Check. At $Q=85$, A’s NPV is $180+4(85-100)=120$ thousand dollars, while B’s is $150+2(85-100)=120$ thousand dollars. The tie agrees with the calculated gap. The common mistake this method avoids is calling the ranking robust just because A leads in the base case.

Use Break-Even Thresholds to Locate a Switch

A break-even threshold for ranking is an assumption value at which the alternatives have equal NPVs. It answers a conditional question: holding the other assumptions fixed, how far must this input move before the ranking changes? It does not say how likely that input value is, nor does it establish what happens when several inputs change together.

When $D$ is linear in one assumption $x$, the threshold can be found by solving $D(x)=0$. If the threshold lies outside the plausible range for $x$, then changes in that input alone do not reverse the ranking within that range. If it lies inside, the ranking is exposed to a switch from that input, though the threshold’s relevance still depends on the other assumptions being held fixed.

Worked Example: Checking Two Assumptions Together

A company compares two hypothetical equipment projects. Their base-case NPVs at $Q=50$ and $C=20$ are $240,000 for A and $216,000 for B. In thousands of dollars, the change in their NPV difference is estimated by $0.8(Q-50)-1.2(C-20)$, where $Q$ is a demand index and $C$ is a variable-cost index. Plausible ranges are $40\le Q\le60$ and $18\le C\le22$. Test whether A remains ahead and find each one-variable break-even threshold at the base value of the other input.

Method. First form the base gap, then evaluate the joint difference over the input rectangle. The gap is $240-216=24$ thousand dollars, so:

$$ D(Q,C)=24+0.8(Q-50)-1.2(C-20). $$

The coefficient of $Q$ is positive, so the gap is smallest at the lowest $Q$. The coefficient of $C$ is negative, so it is smallest at the highest $C$. Thus the worst corner for A is $Q=40,\ C=22$:

$$ D(40,22)=24+0.8(-10)-1.2(2)=24-8-2.4=13.6. $$

The minimum gap is $13,600, which is positive. The other corner, $Q=60,\ C=18$, gives the maximum: $24+0.8(10)-1.2(-2)=24+8+2.4=34.4$. The gap ranges from $13,600 to $34,400, so A remains ahead throughout the stated ranges.

For the demand threshold with cost held at its base value $C=20$, solve $0=24+0.8(Q-50)$. Then $Q-50=-30$, so $Q=20$, outside the stated demand range. For the cost threshold with demand fixed at $Q=50$, solve $0=24-1.2(C-20)$. This gives $C-20=20$, so $C=40$, also outside the stated cost range.

Check. The joint worst-case calculation already shows a positive minimum gap of 13.6 thousand dollars; the thresholds are consistent with A staying ahead on the specified ranges. The common mistake avoided is testing each input only at its base value and overlooking a combined adverse movement.

Preserve Shared Assumptions in the Comparison

When both projects depend on the same uncertain input, that input must take the same value in both NPVs in any valid comparison. Treating the projects’ values as if they could experience different values of a shared driver can create combinations that cannot occur. This can make a ranking seem less stable than it is.

The reason to calculate the difference directly is algebraic: if the same input affects both projects, its effect on the difference is the change in A’s NPV minus the change in B’s NPV. Common effects may cancel fully or partly. Therefore, separate NPV ranges can overlap even when the ranking never changes under the shared assumption. Conversely, a positive base gap may be erased if the projects respond differently to the shared input.

Worked Example: Avoiding Impossible Independent Extremes

Two hypothetical projects have base NPVs of $120,000 for A and $110,000 for B at a shared operating index $X=0$. For every one-unit increase in $X$, each project’s NPV rises by $2,000. The plausible range is $-10\le X\le10$. Does A remain ahead?

Method. Model both NPVs using the same $X$, and subtract. In thousands of dollars, $NPV_A=120+2X$ and $NPV_B=110+2X$. Hence:

$$ D(X)=(120+2X)-(110+2X)=10. $$

The gap is always $10,000, so A remains ahead for every feasible value of the shared index. At $X=-10$, the NPVs are $100,000 and $90,000; at $X=10$, they are $140,000 and $130,000. Both endpoints confirm the same gap.

A mistaken independent-extremes calculation would compare A’s minimum, $100,000, with B’s maximum, $130,000, and claim that a reversal is possible. That calculation pairs A at $X=-10$ with B at $X=10$, even though $X$ is one shared operating condition and cannot take both values at once. The correction is to preserve the shared assumption and calculate the gap using the same $X$.

Use Scenarios to Test Coherent Outcomes

A scenario is a specified combination of assumptions, such as a particular demand environment together with the costs associated with it. Comparing project NPVs within each scenario shows whether the ranking changes across those combinations. The scenarios should preserve relationships among inputs; combining an optimistic demand assumption with a cost outcome that belongs only to a different operating environment may not describe a coherent case.

A ranking reversal in at least one plausible scenario is evidence that the ranking is sensitive to the scenario set. If scenario probabilities are supplied, they can also be used to calculate expected NPVs. That expected ranking answers a different question: which alternative has the higher probability-weighted NPV? It does not mean that the same alternative wins in every scenario.

Worked Example: Distinguishing Scenario Rankings From Expected NPV

A hypothetical firm compares two projects under three mutually exclusive market scenarios. Expansion has probability 0.2, normal conditions probability 0.5, and contraction probability 0.3. The scenario NPVs, in thousands of dollars, are:

Scenario	Probability	Project A NPV	Project B NPV
Expansion	0.2	260	220
Normal	0.5	180	170
Contraction	0.3	70	100

Method. First compare the NPVs scenario by scenario. In expansion, $D=260-220=40$. In normal conditions, $D=180-170=10$. In contraction, $D=70-100=-30$. A leads in two scenarios and B leads in contraction, so the ranking is not stable across these scenarios.

Now calculate the probability-weighted expected NPVs. For A:

$$ E[NPV_A]=0.2(260)+0.5(180)+0.3(70)=52+90+21=163. $$

For B:

$$ E[NPV_B]=0.2(220)+0.5(170)+0.3(100)=44+85+30=159. $$

The expected NPV gap is $163-159=4$ thousand dollars, so A has the higher expected NPV by $4,000. This does not erase the contraction scenario’s $30,000 disadvantage for A. The expected-NPV ranking and the scenario-by-scenario stability conclusion are both correct, but they answer different questions.

Check. The probabilities sum to $0.2+0.5+0.3=1.0$, and the weighted scenario gaps are $0.2(40)+0.5(10)+0.3(-30)=8+5-9=4$, agreeing with the expected NPV gap. The common mistake avoided is treating the higher expected NPV as proof that A ranks first in every outcome.

Practical Interpretation and Common Traps

A useful stability statement identifies the alternatives, the assumptions varied, the tested range or scenarios, and the result for the NPV gap. For example, “A remains ahead of B over the stated demand and cost ranges, with a minimum gap of $13,600” is more informative than “A is the better project.” A conclusion should not imply certainty outside the tested conditions.

Confusing a positive base gap with a robust ranking. A base-case winner may lose when assumptions move. Test the gap over the plausible range.
Ignoring shared drivers. Use the same value of a common assumption for both projects. Do not combine incompatible extremes.
Overlooking interactions. One-variable thresholds hold other assumptions fixed. They do not automatically cover simultaneous changes; evaluate joint combinations or coherent scenarios.
Calling every overlap a reversal. Separate project NPV ranges can overlap while the ranking remains fixed if their extremes rely on different values of a shared input. Compare the difference directly.
Overstating what scenarios establish. A ranking is robust only over the scenarios tested, not over every possible future outcome.

Key takeaway: Assess ranking stability by tracking $NPV_A-NPV_B$, not by relying on the base-case order alone. A strictly positive minimum gap supports A’s ranking over the tested assumption set; a zero or sign change signals a tie or possible reversal that should be reported clearly.

Check Your Understanding

For each problem, compare the NPV difference rather than relying only on the base-case winner.

Projects A and B have base NPVs of $95,000 and $83,000. A’s NPV changes by $1,500 and B’s by $500 per unit of a shared index $Z$, relative to $Z=10$. The plausible range is $4\le Z\le14$. Does A remain strictly ahead? (Answer: Yes; the base gap is $12,000 and the gap changes by $1,000 per unit, so its minimum is $12,000+1,000(4-10)=\$6,000$.)
The NPV difference $D=NPV_A-NPV_B$, in thousands of dollars, is $18+0.6(S-40)$, where $S$ is a sales index. Find the ranking break-even value and state whether the ranking switches over $30\le S\le60$. (Answer: $S=10$; A remains ahead throughout the range because $D(30)=12>0$.)
Three scenarios have probabilities 0.25, 0.50, and 0.25. Their NPV differences $D$, in thousands of dollars, are 24, 6, and $-16$. Which project has the higher expected NPV, and is its ranking stable across scenarios? (Answer: A has the higher expected NPV by $0.25(24)+0.50(6)+0.25(-16)=5$ thousand dollars, but its ranking is not stable because B leads in the third scenario.)
For two alternatives, $D=30+2(U-5)-3(V-4)$, in thousands of dollars. The ranges are $2\le U\le8$ and $3\le V\le5$. Find the minimum gap and assess whether the ranking is robust. (Answer: The minimum is at $U=2,V=5$: $30+2(-3)-3(1)=21$ thousand dollars. A remains strictly ahead throughout the ranges.)

What You'll Learn