On this page
Ho–Lee and Black–Derman–Toy (BDT) are short-rate trees used to value future cash flows under a risk-neutral, or Q, measure. Ho–Lee moves rates by additive amounts and can produce negative nodes; BDT spaces positive rates multiplicatively. For a CAIA® Level II candidate, the practical task is to read the nodes, discount expected future values backwards, and distinguish a calibrated pricing probability from a forecast of where rates will go.
What the two trees are solving
An interest-rate tree represents possible short rates at successive dates. A short rate is the rate used to discount cash flows over one step of the tree, not automatically the yield on every bond. At each node, a security has a value determined by its next cash flow and its values in the next possible states. The modeller works from the final payoffs back to today. This is useful when a bond or an option has cash flows that depend on the path of rates, because one single discount rate cannot describe every future state.
The distinction between a P measure and a Q measure matters before any arithmetic. A P probability is intended to describe real-world likelihood; a Q probability makes discounted traded prices internally consistent within the model. Neither a branch labelled one half nor a higher upper-node rate is evidence that rates have a one-half real-world chance of rising. A candidate should ask which measure a question uses before treating a node probability as an input. CAIA Association places modelling and fixed-income models in its Level II curriculum overview; the official curriculum guidance remains the source for the particular exam year.
How Ho–Lee and BDT differ
A Ho–Lee lattice applies changes to the rate itself. A simple discrete construction can put the next short rate an equal distance above or below a level set for that date. The gap is therefore an amount, often expressed in percentage points. This additive structure is easy to inspect: if the centre changes, the entire row of nodes shifts. Nothing in the additive rule prevents a sufficiently low node from crossing below zero. That possibility is a model property, not a claim that a particular market rate will become negative.
BDT instead constructs a tree with proportional spacing. At one date, a higher node may be a fixed multiple of its lower neighbour. Such a tree keeps rates positive when it starts with positive nodes and uses positive multipliers. The number of percentage points between neighbours gets larger as the rate level gets larger, even if their ratio stays the same. In an exam answer, a constant difference signals an additive construction; a constant ratio signals a multiplicative construction. Do not describe the ratio as a probability or the absolute difference as volatility without checking how the question defines its inputs.
Both trees require calibration rather than an arbitrary choice of every future node. A fitted curve is one that reprices the market instruments selected as calibration targets within the model. A model can be calibrated to observed bond prices and still have limitations for a different payoff or a different date. BDT also uses volatility information to set the spread of future rates; changing that spread can alter the value of an option even when the central level is unchanged. Calibration matches selected present inputs. It does not turn the tree into a prediction of future realised rates or guarantee that all instruments are priced correctly.
A small illustrative backward-induction calculation
Consider an invented two-period zero-coupon claim paying 100 at the end of the second period. Let today’s one-period short rate be 4%. At the next date, suppose an illustrative Ho–Lee-style pair of short-rate nodes is 6% and 2%, equally weighted under Q for this calculation. Treat each period as one year and use simple annual discounting. These are deliberately invented figures for learning the mechanics; they are not a calibrated market curve or a quoted CAIA question. The next-date value is different at the two nodes because the final 100 is discounted at the node-specific short rate.
The upper-rate node gives about 94.34, while the lower-rate node gives about 98.04. Their Q-weighted average is about 96.19. Discount that average once more using today’s 4% short rate; the present value is approximately 92.49. The calculation uses each rate over its own step. Discounting both future node values by today’s rate again, or discounting the final 100 twice at 6%, would attach the wrong rate to at least one period. The value changes if the payoff itself depends on the path, but the backwards sequence remains the same.
For contrast, imagine a BDT-style next row with 3% and 6% nodes. The higher rate is twice the lower rate, whereas the gap is three percentage points. These nodes are not offered as a calibration to the preceding Ho–Lee example: a different pair generally implies a different model price unless the central level is fitted to the same market targets. The useful diagnostic is structural. If the question changes a ratio and asks about the shape of the tree, analyse multiplicative spacing first. If it supplies an actual market zero price, solve the missing level so that the discounted tree value equals that price.
How to read a calibration question
Separate the information into three columns in your notes: observed instruments, model parameters, and target security. Observed zero-coupon prices anchor discounting across maturities; a quoted volatility input constrains how widely rates spread in a model that uses it; the target is the claim you are trying to value after the tree is fitted. Mixing the target claim into the calibration set can make a circular solution look like a valid independent valuation. If a market quote is absent, present your result as an illustrative tree price, as this example does, rather than asserting a market-consistent price.
Work from the payoff backwards one interval at a time. At a node, determine the next-state values, apply the question’s Q branch weights, add any cash flow paid at that node at the correct time, and discount over that interval at the node’s short rate. For a callable bond, exercise rules are applied at the relevant node after continuation value is found; for a plain zero, there is no exercise decision. Make the timing visible. A visually plausible numerical answer can still be wrong if a coupon is counted at the wrong date or if a rate for one row is used on a different row.
Mistakes that change the answer
- Q versus P: a convenient one-half pricing weight does not claim equal real-world likelihoods. The wording of the question determines which probability has a forecasting interpretation.
- Level versus spread: moving the centre of a row and widening the row have different pricing effects, especially for nonlinear payoffs. Keep their roles separate.
- Rate versus yield: each node is a one-period discount rate in the construction. A multi-period yield is inferred from a bond price after discounting through all relevant states.
- Fit versus forecast: a calibrated tree reproduces its chosen current inputs; the future path is still uncertain and the model remains an approximation.
- Additive versus proportional: inspect whether adjacent nodes differ by an amount or by a multiplier before applying a memorised rule.
A good final check is to value the simplest claim in the tree before tackling an option. A zero-coupon payoff provides a clean test of the rate timing and discount convention. Then compare the resulting today value with any observed zero price supplied by the question. If they disagree, revisit calibration or compounding rather than explaining the mismatch away. The CAIA program page gives the wider program context, while the methodology page explains Omni’s approach to study material. For exam specifics, use CAIA Association’s current-year curriculum rather than carrying a prior year’s learning objective forward by assumption.
Sources
- Explore the Curriculum | CAIA Association · caia.org · Retrieved
- Curriculum and Study Tools | CAIA Association · caia.org · Retrieved
- CAIA Charter | CAIA Association · caia.org · Retrieved



