CFA Level II Exam · The Arbitrage-Free Valuation Framework

Pathwise Valuation and Monte Carlo Simulation for Bonds

Updated 7 October 2026 · Fact-checked

Pathwise valuation values a bond by discounting its cash flows along each possible interest rate path, using the one-period rates on that path, then averaging the path values. Monte Carlo simulation generates many random paths from a rate model, so the same method works when a tree is impractical.

Understand Pathwise Valuation and Monte Carlo Simulation

A binomial tree gives you a bond value by working backward node by node. Pathwise valuation reaches the same value another way. It follows each route the short rate can take from today to maturity, values the bond along that route, and then averages.

On one path, you know the one-period rate for every period. You discount each cash flow back using the product of (1 + rate) for every period up to that cash flow. That gives the path value. In a tree with equal probabilities, the bond value is the simple average of all path values. A bond with cash flows at dates 1, 2 and 3 has rates at dates 0, 1 and 2, so there are 2 × 2 = 4 paths.

A tree gets unwieldy because the number of paths grows fast. It also assumes the tree recombines, so the value at a node does not depend on how you got there. That breaks down for path-dependent cash flows. Mortgage-backed securities are the standard case: prepayments depend on how rates have moved before, not only on where they are now.

Monte Carlo simulation solves this. You use a term structure model to draw a large number of random rate paths. You value the bond along each one, using whatever path-dependent cash flows apply. You then average. The model must be arbitrage-free: it is calibrated so that the average simulated value of the benchmark bonds equals their market prices. This is often done by adding a drift adjustment to the simulated rates.

The result is an estimate, not an exact answer. The sampling error shrinks roughly with the square root of the number of paths. To halve it, you need about four times as many paths. You can also add a constant spread to every rate on every path until the average equals the market price. That spread is the option-adjusted spread (OAS).

Key formulas to remember

Path value: Path value = Σ [ CFt ÷ ((1 + r0)(1 + r1)...(1 + rt-1)) ], where r0, r1, ... are the one-period rates on that path; Use only the rates on that path. The rate for each period is the one at the node you are in at the start of that period.
Pathwise bond value: Value = (1 ÷ N) × Σ path values, over N equally likely paths; With equal probabilities this equals the backward-induction value from the same tree. If paths have unequal probabilities, use a probability-weighted average.
Number of paths in a binomial tree: Number of paths = 2^(n − 1) for a bond with n annual cash flow dates; A 3-year bond has rates at dates 0, 1, 2, so 2² = 4 paths.
Monte Carlo sampling error: Standard error ∝ 1 ÷ √(number of paths); Four times as many paths roughly halves the error. This is a proportionality rule, not an exact figure.
Option-adjusted spread (OAS): OAS is the constant spread added to every rate on every path so that average path value = market price; If the model value exceeds the market price, the OAS is positive.

How to solve Pathwise Valuation and Monte Carlo Simulation questions

Use this method for any pathwise valuation question, whether it gives you a small tree or describes a simulation.

1Read the vignette and list the bond's cash flows by date, including the final coupon plus principal.
2Find the one-period rate at each node in the exhibit, and note the probabilities (usually 50% each).
3Write out each path as a sequence of rates, for example r0, then up or down at date 1, and so on.
4For each path, discount every cash flow using the cumulative product of (1 + rate) for the rates on that path only.
5Add the discounted cash flows to get the path value.
6Average the path values using the path probabilities.
7Check with backward induction if time allows. The two answers should match for a recombining tree.
8For a Monte Carlo question, compare the average of the simulated values with the market price to find the OAS direction, and use the number of paths to judge accuracy.

Quickest way: Average the path values, then sanity-check

When to use it: Use it when the tree has two or three periods and the question asks for the bond value or the value on one path.

Compute the discount factors for the first path once: 1 ÷ (1 + r0), then 1 ÷ ((1 + r0)(1 + r1)).
Reuse the date 1 discount factor for every path, since r0 is common to all paths.
Only the later factors change between paths, so recompute just those.
Average the path values and round only at the end.
Quick check: the lower-rate path must have the higher value, and the average must sit between the two.

Common mistakes in Pathwise Valuation and Monte Carlo Simulation

Discounting every cash flow at the same path rate, such as the date 1 rate, for all periods.
Students treat a path like a flat yield curve.
Fix: Discount each period with its own rate. The date t cash flow uses the product of (1 + rate) for all periods up to t.
Using the rate from a different path or node when discounting.
The tree exhibit has many rates and it is easy to pick the wrong node.
Fix: Write each path as a sequence of rates first, then discount along that sequence only.
Forgetting that the coupon at an earlier date is discounted with fewer rates.
Students discount all cash flows with the full path product.
Fix: The date 1 coupon uses only r0. The date 2 cash flow uses r0 and r1.
Miscounting the number of paths.
Students use 2 raised to the number of years instead of the number of rate dates.
Fix: Count the rate dates, which is one less than the number of cash flow dates, then take 2 to that power.
Treating Monte Carlo results as exact and assuming more paths removes all error.
The output looks precise.
Fix: Remember it is a sample estimate. More paths reduce the error roughly with the square root of the count, but never to zero.
Saying Monte Carlo is needed only because trees are too slow.
Students miss the path-dependence point.
Fix: The main reason is path-dependent cash flows, such as mortgage prepayments, which a recombining tree cannot capture. Multiple rate factors are another reason.

Worked examples

Example 1

A 2-year, 6% annual-coupon bond has face value 100. The current one-period rate is 3.00%. At date 1 the rate is 5.00% in the up state and 4.00% in the down state, each with probability 50%. (1) What is the value on the up path? (2) What is the value on the down path? (3) What is the bond's value using pathwise valuation? Options for (3): A. 103.84, B. 104.31, C. 104.78.

Show the solution

Cash flows: 6 at date 1 and 106 at date 2. The date 1 coupon is discounted at 3.00% on both paths: 6 ÷ 1.03 = 5.8252.
Up path: the date 2 cash flow is discounted by 1.03 × 1.05 = 1.0815. 106 ÷ 1.0815 = 98.0120. Path value = 5.8252 + 98.0120 = 103.8372.
Down path: the discount factor is 1.03 × 1.04 = 1.0712. 106 ÷ 1.0712 = 98.9545. Path value = 5.8252 + 98.9545 = 104.7797.
Average: (103.8372 + 104.7797) ÷ 2 = 104.3085, which is 104.31.
Check by backward induction: up node = 106 ÷ 1.05 + 6 = 106.9524. Down node = 106 ÷ 1.04 + 6 = 107.9231. Average = 107.4377. Discount at 1.03: 104.3085. It matches.

Answer: (1) 103.84. (2) 104.78. (3) B, 104.31. The down path has the higher value because lower rates mean less discounting.

Example 2

An analyst values a mortgage-backed security with Monte Carlo simulation using five rate paths from an arbitrage-free model. The path values are 101.2, 99.8, 100.6, 98.9 and 100.5. The market price is 99.50. (1) What is the model value? (2) Is the OAS positive or negative? (3) To roughly halve the sampling error, how many paths are needed if the analyst now uses 5,000? Options for (3): A. 10,000, B. 20,000, C. 2,500.

Show the solution

Sum the path values: 101.2 + 99.8 + 100.6 + 98.9 + 100.5 = 501.0.
Model value = 501.0 ÷ 5 = 100.2.
The model value exceeds the market price of 99.50. A spread added to every rate on every path lowers the path values, so a positive spread is needed to bring the average down to 99.50. The OAS is positive.
Sampling error is proportional to 1 ÷ √N. To halve it, √N must double, so N must be four times as large: 4 × 5,000 = 20,000.

Answer: (1) 100.2. (2) Positive. (3) B, 20,000.

Exam tips

Write each path as a sequence of rates before you discount. This one habit prevents most errors.
If a question gives a tree and asks for the bond value, backward induction is often faster. Use pathwise only when the question asks for path values.
Expect conceptual questions on Monte Carlo versus binomial: path-dependent cash flows, multiple factors, calibration to market prices, and that results are estimates.
For an OAS question, compare the model value with the market price. A higher model value means a positive OAS.
Read the vignette for the number of paths and the probabilities. Do not assume equal probabilities unless the vignette says so.

Pathwise Valuation and Monte Carlo Simulation in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Pathwise Valuation and Monte Carlo Simulation: frequently asked questions

What is pathwise valuation in CFA Level II?

It is a way of valuing a bond by discounting its cash flows along each possible interest rate path and averaging the path values. With equal probabilities in a recombining tree, it gives the same value as backward induction. It is most useful when cash flows depend on the path.

How is Monte Carlo different from the binomial tree?

A binomial tree uses a limited set of recombining nodes, so value at a node does not depend on how you got there. Monte Carlo draws many random paths from a rate model, so it handles path-dependent cash flows and several factors. The cost is more computation and an answer that is an estimate.

Why must the simulated paths be arbitrage-free?

The paths are calibrated so that the average simulated value of benchmark bonds equals their market prices. If they were not, the model would misprice known securities, and its value for a complex bond could not be trusted.

How many paths are in a binomial pathwise valuation?

For a bond with n annual cash flow dates there are rates at n − 1 later dates, giving 2^(n − 1) paths. A 3-year bond has 4 paths. Always count rate dates, not years.