Start From the Market, Not From Theory: the Ho-Lee Model

There is a difference between a model that tells you what the world should look like and a model that tells you what to charge. Vasicek and CIR were the first kind. Ho and Lee built the first of the second kind, and in doing so they created the design pattern that every practical interest rate model has followed since.

The problem: your model disagrees with the screen

Imagine you are a trader in 1985. A client wants to buy an option on a Treasury bond. You reach for the best available theory, Vasicek's model, calibrate it as well as you can, and it tells you the five-year yield ought to be 8.4 percent.

The screen says 8.9 percent.

Now what? You cannot trade on your model, because your model is mispricing the underlying bond before you even get to the option. If the model cannot reproduce the price of a plain vanilla bond that anyone can see quoted, why on earth would you trust it on a derivative?

This is not a calibration failure you can fix by trying harder. It is structural. Equilibrium models like Vasicek and CIR have a handful of parameters, and a real yield curve has dozens of points with all sorts of kinks and humps in it. A few parameters will never thread that needle exactly. The model produces a yield curve, and it is the wrong one.

Ho and Lee's insight was that for pricing derivatives, you do not want a model that produces the yield curve. You want a model that swallows it.

The key idea via analogy: the map is given, only the roads are yours

Think of two ways to draw a map of a mountain range.

The equilibrium modeller says: here are the laws of geology, here is how tectonic plates behave, let me derive where the mountains must be. They produce a beautiful, principled map. It does not match the actual mountains.

Ho and Lee say: forget it. The mountains are right there. I will trace them. Today's yield curve is not something to be explained, it is data, and I will take it as given, exactly, point for point. My job is not to say where rates should be. My job is to say where they can go from here, and to make sure that in doing so, I never accidentally create a free lunch.

Mechanically, they build a binomial tree. At each step, the whole yield curve either moves "up" or "down" by some amount. Starting from today's observed curve, the tree branches out into the future. So far this is just bookkeeping. The clever part is the constraint they impose.

At every single node of that tree, the model must be arbitrage free: no combination of bonds bought and sold at that node can be turned into a guaranteed profit. Ho and Lee show that this requirement, applied everywhere, is enormously restrictive. It does not merely rule out some silly trees. It pins down almost the entire structure. Once you have fixed today's curve and chosen how volatile rates are, the no-arbitrage condition determines everything else, including where the drift of the tree must go.

That is the deep result and it is worth pausing on. The drift is not a free parameter you get to choose. It is forced on you. The model must lean in exactly the direction needed to keep today's observed curve consistent with tomorrow's possibilities. If today's curve is steeply upward-sloping, the tree is forced to drift rates upward, because that is the only way the observed long bond price can be right without an arbitrage existing.

With the tree built, pricing anything is mechanical: bond options, callable bonds, whatever. You roll backwards through the tree. And by construction, the model reproduces every bond price you can see on the screen, perfectly, because you fed them in.

Why it mattered

• It invented the calibrate-to-the-curve paradigm. This is the whole game now. Hull-White, Black-Derman-Toy, Black-Karasinski, Heath-Jarrow-Morton and the LIBOR market model are all, at bottom, doing what Ho and Lee did: taking the observed curve as an input and letting no-arbitrage do the rest. Ho and Lee got there first.
• It made interest rate derivatives priceable. Before this, pricing a bond option meant living with a model that could not price the bond. After this, you could quote a callable bond or a bond option with a straight face.
• It showed no-arbitrage has teeth. The realisation that "no free lunch, everywhere, at every node" is enough to determine the drift of the entire system was a genuine surprise, and it is the seed that Heath, Jarrow and Morton grew into a general theory six years later.
• It is beautifully simple to implement. A binomial tree is something you can build in a spreadsheet. That accessibility is a big part of why the idea spread so fast.

The honest limitations

The Ho-Lee model is, frankly, a bad model of interest rates. That is fine, it was a proof of concept, but the flaws are real and they are the reason nobody uses it in its original form.

• No mean reversion. Rates in the Ho-Lee world just wander. There is no elastic lead pulling them back to a normal level. In practice this means the model thinks a rate of 40 percent, thirty years from now, is perfectly plausible. It is not. This is the flaw Hull and White fixed, essentially by grafting Vasicek's mean reversion onto Ho and Lee's calibration trick.
• Rates can go negative. The tree moves rates up or down by fixed amounts regardless of where they are, so they can wander straight through zero. In 1986 this was considered a fatal defect. In the 2010s it stopped being one, but the model still gives no reason for rates to hover near zero.
• Constant volatility. Every rate, at every maturity, at every point in the future, jiggles by the same amount. Real rate volatility is nothing like that: short rates and long rates have very different volatilities, and volatility itself changes over time.
• Only parallel shifts. In this model, when rates move, the entire curve moves up or down together. It cannot steepen, flatten, twist or hump. Since roughly a third of the real variation in bond returns comes from the curve changing shape rather than level (see Litterman and Scheinkman), this is a serious restriction for anything whose payoff depends on the spread between maturities.
• Perfect fit is a double-edged sword. Fitting today's curve exactly sounds like a virtue, and for pricing it is. But it also means the model absorbs whatever noise, illiquidity and quirks are in today's quotes without comment. A misprinted quote becomes gospel. And a model calibrated to fit everything today tells you nothing about whether today's curve is cheap or expensive, which is a question a relative-value trader very much wants answered.

The one-line takeaway

Ho and Lee stopped trying to explain the yield curve and started taking it as given, showing that once you fix today's curve and refuse to allow free lunches anywhere in the future, the rest of the model builds itself, an inversion that every practical interest rate model has copied ever since.