Newton’s Method from the NPV Derivative

Why Use the NPV Derivative to Estimate IRR?

An internal rate of return is a rate at which a project’s net present value is zero. For a simple cash-flow pattern, the equation may be solved directly; for uneven cash flows, it is often more practical to estimate the root of the NPV function. Newton’s method uses both the current NPV and the rate at which NPV is changing to choose a new estimate.

The method can converge quickly when the current estimate is reasonably close to a simple root. It is not guaranteed to do so: if the NPV derivative is small, the method can take an exceptionally large step, possibly to a rate outside the permissible domain. Every estimate should therefore be checked in the original NPV equation.

Definition: Suppose cash flow $C_t$ occurs at each integer time $t=0,1,\ldots,n$, with inflows positive and outflows negative. Under annual effective rate $i>-1$, define $$ \operatorname{NPV}(i)=\sum_{t=0}^{n} C_t(1+i)^{-t}. $$ An IRR is a rate $i>-1$ satisfying $\operatorname{NPV}(i)=0$. The cash-flow timing and the rate period must agree: if periods are years, $i$ is an effective annual rate.

Deriving the Newton Update

For a fixed set of cash flows, differentiate each discounted term with respect to the rate. The time-zero cash flow has derivative zero, while a payment at time $t>0$ contributes a factor of $-t$ from differentiating $(1+i)^{-t}$.

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

To justify the update, approximate the NPV function near a current estimate $i_k$ by its tangent line. At the next estimate, set this linear approximation equal to zero:

0\approx \operatorname{NPV}(i_k) +\operatorname{NPV}'(i_k)(i_{k+1}-i_k).

Solving for $i_{k+1}$ gives Newton’s method:

Theorem: If $\operatorname{NPV}'(i_k)\ne 0$, the Newton update for estimating an IRR is $$ i_{k+1}=i_k-\frac{\operatorname{NPV}(i_k)}{\operatorname{NPV}'(i_k)}. $$ This is a tangent-line estimate, not a guarantee that the new rate is an IRR or even lies in the valid domain $i>-1$. The method is generally most dependable near a root where the function is differentiable and its derivative is not zero.

The derivative has units of currency per unit change in the rate, so the ratio $\operatorname{NPV}(i_k)/\operatorname{NPV}'(i_k)$ is a rate change. A positive NPV with a negative derivative produces an upward rate update, as expected for a conventional investment whose NPV declines as its discount rate rises.

Applying Newton’s Method

Write the NPV function.
Include every cash flow at its correct time and use the same period for the rate and timing.

Differentiate.
For a cash flow at time $t$, its derivative term includes the multiplier $-t$.

Update the estimate.
Substitute the current rate, NPV, and derivative into the Newton formula.

Check the result.
Substitute the new rate back into NPV. A small residual supports convergence; also confirm that the rate is in the domain.

Worked Example: Two Newton Updates for a Two-Year Project

A hypothetical project costs $1,000 now and returns $1,440 at the end of year 2, with no other cash flows. Use Newton’s method starting at $i_0=0.100000$. Rates are effective annual rates. Find two successive estimates and check the second one.

Method and NPV. With $C_0=-1000$, $C_2=1440$, and $C_1=0$, $$ \operatorname{NPV}(i)=-1000+\frac{1440}{(1+i)^2}, \qquad \operatorname{NPV}'(i)=-\frac{2880}{(1+i)^3}. $$

At $i_0=0.100000$, $$ \operatorname{NPV}(i_0)=-1000+\frac{1440}{1.1^2} =-1000+1190.082645=190.082645, $$ and $$ \operatorname{NPV}'(i_0)=-\frac{2880}{1.1^3} =-2163.7866. $$ Thus $$ i_1=0.100000-\frac{190.082645}{-2163.7866} \approx 0.187847. $$

Second update. Using the unrounded first estimate in the calculations, $$ \operatorname{NPV}(i_1)\approx20.566, \qquad \operatorname{NPV}'(i_1)\approx-1718.3. $$ Therefore, $$ i_2=i_1-\frac{\operatorname{NPV}(i_1)}{\operatorname{NPV}'(i_1)} \approx0.187847+\frac{20.566}{1718.3} \approx0.199816. $$

Check. At $i_2$, $\operatorname{NPV}(i_2)\approx0.31$, much closer to zero than the initial NPV of $190.08. In fact, the equation has the exact solution $1+i=\sqrt{1440/1000}=1.2$, so the IRR is $0.200000$. The second estimate is close to that root.

Common mistake avoided. The derivative is not $-1440/(1+i)^3$: differentiating the year-2 term produces the time multiplier 2, giving $-2880/(1+i)^3$. Omitting that multiplier changes every Newton step.

Worked Example: Uneven Conventional Cash Flows

A hypothetical investment has cash flows of $-$500$ at time 0, $+$300$ at time 1, and $+$300$ at time 2. Starting at $i_0=0.100000$, calculate two Newton estimates and check the second. Use annual periods.

Method. The NPV and its derivative are $$ \operatorname{NPV}(i)=-500+\frac{300}{1+i}+\frac{300}{(1+i)^2}, \qquad \operatorname{NPV}'(i)=-\frac{300}{(1+i)^2}-\frac{600}{(1+i)^3}. $$ At $i_0=0.10$, $$ \operatorname{NPV}(i_0)=-500+\frac{300}{1.1}+\frac{300}{1.1^2} =-500+272.727273+247.933884 =20.661157, $$ and $$ \operatorname{NPV}'(i_0) =-\frac{300}{1.21}-\frac{600}{1.331} =-247.933884-450.789031 =-698.722915. $$

The first update is $$ i_1=0.100000-\frac{20.661157}{-698.722915} \approx0.129570. $$ Substituting this estimate into the NPV and derivative gives approximately $$ \operatorname{NPV}(i_1)\approx0.711, \qquad \operatorname{NPV}'(i_1)\approx-651.4. $$ The second update is consequently $$ i_2\approx0.129570-\frac{0.711}{-651.4} \approx0.130661. $$

Check. Substitution at $i_2$ gives an NPV of approximately $0.001, close to zero. As an independent check, setting $y=1+i$ in the NPV equation gives $5y^2-3y-3=0$. Its positive solution is $y=(3+\sqrt{69})/10$, so $i\approx0.130662$, consistent with the Newton estimate.

Common mistake avoided. The year-2 derivative term is $-2(300)/(1+i)^3$, not $-300/(1+i)^3$. The time multiplier is essential even when the cash flow itself is level across periods.

When a Small Derivative Makes the Update Unstable

The update divides by the derivative. If $\operatorname{NPV}'(i_k)$ is close to zero while the NPV is not, the quotient can be very large. The tangent line is then nearly horizontal and may predict a zero far from the current estimate. A small NPV derivative is therefore a warning to inspect the step, not a reason to accept it automatically.

Worked Example: A Long-Dated Cash Flow Produces a Huge Step

A hypothetical project pays $-$1,000$ now and $+$2,000$ at the end of year 20. Starting from $i_0=1.00$ (100% effective annually), calculate one Newton update and assess it. The rate domain is $i>-1$.

Method. The NPV and derivative are $$ \operatorname{NPV}(i)=-1000+\frac{2000}{(1+i)^{20}}, \qquad \operatorname{NPV}'(i)=-\frac{40000}{(1+i)^{21}}. $$ At $i_0=1$, $$ \operatorname{NPV}(1)=-1000+\frac{2000}{2^{20}} =-999.998093, $$ while $$ \operatorname{NPV}'(1)=-\frac{40000}{2^{21}} \approx-0.0190735. $$ The resulting update is $$ i_1 =1-\frac{-999.998093}{-0.0190735} \approx-52427.7. $$

Check and interpretation. The update is below $-1$, outside the domain in which the annual effective discount factor $1/(1+i)$ is valid. It is not a usable IRR estimate. The equation itself has a valid positive IRR, since $(1+i)^{20}=2$, giving $i=2^{1/20}-1\approx0.03527$. The failure here is the enormous step from a distant initial rate, caused by the small derivative at that rate.

Common mistake avoided. A Newton update is not automatically admissible just because it follows the formula. Check the new rate’s domain and evaluate NPV before treating it as an improved estimate.

Practical Checks and Exam Traps

For conventional cash flows—one initial outflow followed by positive inflows—the derivative is negative for every $i>-1$, because every nonzero post-investment term in the derivative is negative. Thus, if NPV is positive, the Newton update raises the rate; if NPV is negative, it lowers the rate. These sign checks can reveal an arithmetic or sign error. They do not establish that a large step is safe.

The derivative’s time multiplier is a frequent source of error. For a payment $C_t$ at time $t$, differentiate its own term: $$ \frac{d}{di}\left(C_t(1+i)^{-t}\right) =-tC_t(1+i)^{-t-1}. $$ The exponent in the discounted value becomes $t+1$, and the multiplier $t$ remains. The time-zero term has no derivative. Also keep full precision during the iterations; rounding a rate too early can alter later estimates, especially when the derivative is small.

Newton’s method is useful because it can improve an estimate rapidly near a root, but it is a local method. A small residual at one candidate rate is evidence that the NPV equation is nearly satisfied, whereas a large step, a zero derivative, or a rate outside $i>-1$ signals that the iteration cannot be used as-is. This derivative-based approach is aligned with the SOA Exam FM (Financial Mathematics) syllabus.

Key takeaway: Calculate $\operatorname{NPV}(i_k)$ and $\operatorname{NPV}'(i_k)$ from the same dated cash flows, apply $i_{k+1}=i_k-\operatorname{NPV}(i_k)/\operatorname{NPV}'(i_k)$, then check both the residual and the domain of the new rate.

Check Your Understanding

For each problem, use the stated annual effective rate and show the NPV, derivative, and Newton update before comparing with the answer.

A hypothetical project has cash flows $-$200$ now and $+$240$ in one year. Starting at $i_0=0.10$, find one Newton estimate. (Answer: $i_1\approx0.1917$.)
A project has cash flows $-$500$ now and $+$300$ at each of years 1 and 2. Starting at $i_0=0.10$, find one Newton estimate. (Answer: $i_1\approx0.129570$.)
A hypothetical project has cash flows $-$1,000$ now, $+$400$ at year 1, and $+$800$ at year 2. Starting at $i_0=0.10$, find one Newton estimate. (Answer: $i_1\approx0.116177$.)
A project has cash flows $-$1,000$ now and $+$1,500$ at year 3. Starting at $i_0=0.10$, find one Newton estimate. (Answer: $i_1\approx0.1413$.)
A project has cash flows $-$1,000$ now and $+$2,000$ at year 20. Starting at $i_0=1.00$, find one Newton update and say whether it is in the valid rate domain. (Answer: $i_1\approx-52427.7$, outside $i>-1$.)

What You'll Learn