Choosing IRR Solver Guesses and Diagnosing Failure

Why Solver Setup Matters

An IRR solver returns a numerical answer to an equation; it does not confirm that the equation, rate units, or cash-flow timing were entered correctly. A solver can fail to converge, stop at an implausible estimate, or report a number that is not a root of the intended NPV equation. The essential check is to substitute the reported rate into that equation and evaluate NPV independently.

This tutorial focuses on conventional cash flows: an initial outflow followed by nonnegative inflows, with at least one positive inflow. For equally spaced periods, let $C_t$ be the cash flow at time $t$, where $C_0<0$, $C_t\geq 0$ for $t\geq1$, and at least one later $C_t$ is positive. The effective rate per period is $i$, with $i>-1$.

Definition: For equally spaced cash flows, the NPV at rate $i$ is $$ \operatorname{NPV}(i)=\sum_{t=0}^{n} C_t(1+i)^{-t}. $$ An IRR is a rate $i>-1$ for which $\operatorname{NPV}(i)=0$. A numerical solver estimates such a rate; the NPV equation defines what the estimate must satisfy.

For a conventional pattern, the NPV decreases as the rate increases, and there is one IRR in the domain $i>-1$. Thus, when a solver gives a strange answer for this cash-flow pattern, first investigate the setup and the numerical process rather than assuming there are several conventional IRRs.

Choosing a Starting Rate or Bracket

A starting rate should be a plausible rate per cash-flow period, not an arbitrary number in a convenient-looking format. A useful first step is to evaluate NPV at a few trial rates that make sense for the project. If two trial rates produce opposite NPV signs, they bracket an IRR. The bracket provides a reliable basis for bisection and a useful check for other methods.

If no bracket is available, begin with a moderate rate consistent with the scale and timing of the cash flows. Evaluate NPV there. A positive NPV means the trial rate is below the IRR for conventional cash flows; a negative NPV means it is above the IRR. Adjust the trial rate accordingly, or find a rate on the other side to create a bracket. Do not treat a guess as evidence that a root exists.

Method	Starting information	Practical strength	Setting to check
Bisection	Two rates with opposite NPV signs	Reliable progress while preserving the bracket	Bracket endpoints and stopping tolerance
Newton’s method	One starting rate and the NPV derivative	Can converge quickly from a useful guess	Starting rate, derivative, and whether each step stays above $-1$
Secant method	Two starting rates and their NPVs	Does not require a derivative	Starting rates with distinct NPV values and the resulting steps

A bracket is especially valuable when the initial estimate is uncertain. Newton’s method and the secant method can take steps outside a useful region; a result outside $i>-1$ is not an admissible effective rate for this model. Even a step that remains in the domain can move far from the root. A bracketed method trades some speed for controlled progress.

Why NPV Signs and the Derivative Help

The NPV sign tells us which side of the root a trial rate lies on for conventional cash flows. The derivative explains both that behavior and why Newton’s method can be sensitive to its starting value. Differentiating the finite sum term by term gives:

\operatorname{NPV}'(i) = -\sum_{t=1}^{n} tC_t(1+i)^{-t-1}.

Every term in this sum is nonpositive, and at least one is strictly negative because at least one later cash flow is positive. Therefore $\operatorname{NPV}'(i)<0$ for $i>-1$. NPV is strictly decreasing, so a change from positive to negative brackets the unique root. The same derivative appears in Newton’s update:

i_{\text{new}} = i_{\text{old}} - \frac{\operatorname{NPV}(i_{\text{old}})} {\operatorname{NPV}'(i_{\text{old}})}.

This update is based on the tangent to the NPV curve at the current estimate. If that tangent intersects the rate axis far from the root, the next estimate can be poor. The equation gives a candidate step, not a guarantee that the step is sensible.

Worked Example: Find a Useful Bracket and Check the Root

A project requires $\$1{,}000$ now and returns $\$500$ at the end of year 1 and $\$700$ at the end of year 2. Find a useful bracket for its effective annual IRR, solve the equation, and verify the answer. Use annual periods.

Method. Evaluate NPV at nearby trial rates, use the sign change to bracket the root, then solve the quadratic in $1+i$. This avoids accepting a solver’s estimate without checking its NPV.

The equation is $$ -1000+\frac{500}{1+i}+\frac{700}{(1+i)^2}=0. $$ At $i=10\%$, $$ \operatorname{NPV}(0.10) =-1000+\frac{500}{1.10}+\frac{700}{1.10^2} =-1000+454.5455+578.5124 =\$33.0579. $$ At $i=15\%$, $$ \operatorname{NPV}(0.15) =-1000+\frac{500}{1.15}+\frac{700}{1.15^2} =-1000+434.7826+529.3006 =-\$35.9168. $$ The signs differ, so the IRR is between $10\%$ and $15\%$.

Set $x=1+i$, so $x>0$. Multiplying the NPV equation by $x^2$ gives $$ 1000x^2-500x-700=0. $$ The positive solution is $$ x=\frac{500+\sqrt{3{,}050{,}000}}{2000} =1.12321246, \qquad i=x-1=0.12321246. $$ The negative quadratic solution is inadmissible because $x=1+i$ must be positive. The IRR is $12.3212\%$ to four decimal places.

Check. At the calculated rate, the discounted inflows are approximately $$ \frac{500}{1.12321246}=\$445.15, \qquad \frac{700}{1.12321246^2}=\$554.85. $$ Their sum is $\$1{,}000.00$, so NPV is $\$0.00$ to the nearest cent. The result also lies inside the bracket and has the expected sign change.

Common mistake avoided: using a rate merely because a solver accepts it. Here, the NPV signs establish a bracket before the equation is solved, and substituting the answer checks that it is actually a root.

Diagnosing a Poor Starting Guess

A poor Newton starting value can produce an unexpectedly large step even when the correct IRR is ordinary. This happens because the update divides the current NPV by the derivative. If the starting point is far from the root, the tangent may not approximate the curve well over the distance to the root. The next rate is then a mathematical update, not necessarily a good practical estimate.

Worked Example: Newton’s Method Takes an Unhelpful Step

A project has cash flows of $-\$1{,}000$ at time 0 and $\$600$ at the end of each of years 1 and 2. An analyst starts Newton’s method at $i=100\%$. Diagnose the first steps and identify a better strategy.

Method. Compute NPV and its derivative at the stated guess, apply one Newton update, and compare it with a bracket based on NPV signs. This shows why an accepted solver step should not be confused with a reliable one.

The NPV function and derivative are $$ \operatorname{NPV}(i) =-1000+\frac{600}{1+i}+\frac{600}{(1+i)^2}, \qquad \operatorname{NPV}'(i) =-\frac{600}{(1+i)^2}-\frac{1200}{(1+i)^3}. $$ At $i=1$, $$ \operatorname{NPV}(1)=-1000+300+150=-\$550, \qquad \operatorname{NPV}'(1)=-150-150=-300. $$ Thus the next Newton estimate is $$ i_1=1-\frac{-550}{-300}=-0.833333. $$ This large move takes the estimate from $100\%$ to approximately $-83.3333\%$, although the target rate is not close to that negative rate.

At $i_1=-5/6$, $1+i_1=1/6$, giving $$ \operatorname{NPV}(i_1) =-1000+600(6)+600(36) =\$24{,}200, $$ and $$ \operatorname{NPV}'(i_1) =-600(36)-1200(216) =-280{,}800. $$ The second Newton estimate is $$ i_2=-0.833333-\frac{24{,}200}{-280{,}800} \approx -0.747151. $$ This is still far below the root. The method has not violated $i>-1$, but its steps have been poorly located.

A better strategy is to check a bracket. At $10\%$, NPV is $$ -1000+\frac{600}{1.10}+\frac{600}{1.10^2} =\$41.3223. $$ At $20\%$, NPV is $$ -1000+\frac{600}{1.20}+\frac{600}{1.20^2} =-\$83.3333. $$ The root is therefore between $10\%$ and $20\%$. Solving with $x=1+i$ gives $$ 1000x^2-600x-600=0, \qquad x=\frac{600+\sqrt{2{,}760{,}000}}{2000} =1.13066239. $$ Therefore $i=0.13066239$, or $13.0662\%$ to four decimal places.

Check. At this rate, the discounted inflows are approximately $\$530.66$ and $\$469.34$, totaling $\$1{,}000.00$. NPV is $\$0.00$ to the nearest cent. The bracketed method avoids relying on the distant Newton starting value.

Common mistake avoided: assuming Newton must move toward the root at every iteration. Its tangent-based step can be unhelpful; use a sign-changing bracket and bisection when reliable progress matters.

Checking Units, Residuals, and Solver Settings

Many apparent numerical failures are actually input or interpretation errors. An effective rate of $13\%$ is entered as $0.13$ in a calculation that expects a decimal, not as $13$. A solver might display the same rate as either $0.13$ or $13\%$, so check the software’s convention. Also confirm that the rate is per cash-flow period and that the cash flows have been entered with the intended signs and dates.

A second diagnostic is the NPV residual: calculate $\operatorname{NPV}(\hat{i})$ at the reported estimate $\hat{i}$. A small change between successive rate estimates does not by itself prove that NPV is near zero. Conversely, a rounded displayed rate can have a visible residual even when the solver used more internal digits. Verify using the unrounded estimate when available, and state a sensible tolerance in the units of the cash flows.

Worked Example: Detect a Percent-versus-Decimal Error

A project has cash flows of $-\$5{,}000$ now, $\$3{,}000$ at the end of year 1, and $\$3{,}000$ at the end of year 2. A solver reports “13.” Determine whether that is a credible rate if the solver expects decimal input, and check the intended rate near $13\%$.

Method. Evaluate the original NPV equation at both $i=13$ and $i=0.13$. A root must make NPV zero, not merely resemble a plausible percentage when displayed.

At $i=13$, which is $1300\%$ as an effective annual rate, $$ \operatorname{NPV}(13) =-5000+\frac{3000}{14}+\frac{3000}{196} =-5000+214.2857+15.3061 =-\$4{,}770.4082. $$ This is nowhere near zero, so the displayed “13” cannot be accepted as the decimal-rate IRR.

Now evaluate at $i=0.13$, or $13\%$: $$ \operatorname{NPV}(0.13) =-5000+\frac{3000}{1.13}+\frac{3000}{1.13^2} \approx -5000+2654.8673+2349.4400 =\$4.3073. $$ This small positive NPV indicates that the IRR is slightly above $13\%$, not exactly $13\%$.

For a direct solution, put $x=1+i$: $$ 5000x^2-3000x-3000=0, \qquad x=\frac{3+\sqrt{69}}{10}=1.13066239. $$ Thus $i=0.13066239$, or $13.0662\%$ to four decimal places. At this rate, the discounted inflows are approximately $\$2{,}653.31$ and $\$2{,}346.69$. They total $\$5{,}000.00$, giving NPV $\$0.00$ to the nearest cent.

Common mistake avoided: reading the solver’s displayed number as a percentage without checking its units. The independent NPV calculation exposes the error immediately.

A Practical Diagnostic Sequence

For conventional cash flows, a disciplined check is more useful than repeatedly changing guesses without recording what happened. Use the following sequence when an IRR solver does not return a convincing result.

Reconstruct the timeline.
Confirm every cash-flow amount, sign, and period. The rate must match the spacing of the cash flows.

Check the rate convention.
Determine whether the solver expects a decimal or a percentage, and confirm the rate is per period.

Evaluate trial NPVs.
Use reasonable rates. Opposite signs identify a bracket; for conventional cash flows, NPV decreases with the rate.

Choose a suitable solver approach.
Use bisection for a reliable bracketed solution. Use Newton or secant when guesses are sensible, and monitor the successive estimates.

Verify independently.
Substitute the unrounded estimate into the original NPV equation and report the residual at an appropriate precision.

If the solver stops because the change in rate is small but the NPV residual is not, the estimate has not passed the essential check. Review the model and solver settings, increase the iteration limit if appropriate, or use bisection within a valid bracket. If no opposite-sign bracket can be found for a conventional pattern, recheck the cash flows, rate domain, units, and period spacing before interpreting the output.

Key takeaway: A credible conventional-cash-flow IRR is in the admissible rate domain, lies consistently with the NPV signs, and makes the original NPV approximately zero. Solver output alone is not verification.

Check Your Understanding

For each problem, use the stated periods and rates. Check a numerical answer by considering its NPV or the sign of the NPV at the stated trial rates.

A project pays $\$2{,}400$ now and returns $\$3{,}000$ exactly two years later, with no other cash flows. Find its effective annual IRR. (Answer: $11.8034\%$, to four decimal places.)
A project has cash flows $-\$1{,}800$ now, $\$700$ at year 1, and $\$1{,}300$ at year 2. Evaluate NPV at $5\%$ and $10\%$, and state the bracket indicated by their signs. (Answer: NPV at $5\%$ is approximately $+\$45.81$, NPV at $10\%$ is approximately $-\$89.26$; the IRR is bracketed by $5\%$ and $10\%$.)
A solver expects decimal rates. An analyst enters $9$ for a $9\%$ trial on cash flows $-\$1{,}200$ now, $\$700$ at year 1, and $\$700$ at year 2. What NPV does that input actually evaluate? (Answer: $-\$1{,}123.00$, because $i=9$ means $900\%$, so $-1200+700/10+700/100=-1123$.)
For cash flows $-\$2{,}000$ now, $\$900$ at year 1, and $\$1{,}400$ at year 2, calculate one Newton update from $i=10\%$, using the derivative of NPV. (Answer: $i_{\text{new}}\approx0.091291$, or $9.1291\%$.)
A project has cash flows $-\$5{,}000$ now, $\$2{,}400$ at year 1, and $\$3{,}000$ at year 2. Show that $5\%$ and $6\%$ bracket the IRR, then solve the quadratic in $1+i$. (Answer: the IRR is approximately $5.0925\%$; NPV is positive at $5\%$ and negative at $6\%$.)

What You'll Learn