Backward Induction: How the Binomial Model Actually Prices an Option

There’s a particular moment in the binomial option pricing model where candidates get visibly relieved — the moment they realize the math doesn’t require predicting where the stock price is actually headed. You don’t need a view on the future. You need something almost the opposite: you need to start at the end, where the answer is obvious, and work your way backward to the beginning, where it isn’t.

That technique has a name — backward induction — and once it clicks, the entire binomial framework stops feeling like memorized formula-plugging and starts feeling like something you could rebuild from scratch if you had to.

The Core Idea, Stripped of Notation

Backward induction is a method for solving multi-period problems by starting at the final period, where the outcome is known with certainty, and then working backward one step at a time, using each known outcome to determine the correct value at the step before it.

Why does this work for option pricing specifically? Because an option’s payoff at expiration is completely unambiguous. A call option at expiration is worth exactly max(0, Stock Price − Strike Price) — no judgment calls, no uncertainty, no modeling assumptions needed. You know the terminal payoff with total precision the moment you know the terminal stock price.

The hard part isn’t the end of the tree. It’s everywhere before that — every earlier node where you don’t yet know which direction the stock will move, and yet you still need to assign a value to the option today. Backward induction solves that problem by refusing to guess. Instead, it uses the values you already know with certainty at the end of the tree to derive values at the node just before it, then uses those newly-derived values to derive the node before that, and so on, walking backward one period at a time until you land back at today.

Building the Tree: A Single-Period Example First

Before tackling a multi-period tree, it’s worth nailing the single-period case completely, since every multi-period binomial tree is really just this same logic, repeated.

Suppose a stock currently trades at ₹100. Over the next period, it can move up to ₹120 or down to ₹90 — these are the only two possible outcomes, which is exactly what makes this a binomial model. You’re pricing a call option with a strike price of ₹100, expiring at the end of this single period.

At the up-node, the stock is worth ₹120, so the call payoff is max(0, ₹120 − ₹100) = ₹20.

At the down-node, the stock is worth ₹90, so the call payoff is max(0, ₹90 − ₹100) = ₹0.

Notice something important: these two payoffs required no induction at all. They’re terminal values, known with certainty given each branch. The real question — what’s the option worth today, before you know which branch actually happens — is where backward induction earns its keep.

Working Backward: The Replicating Portfolio Logic

Here’s where candidates sometimes want to just average the two payoffs, weighted by some assumed probability of an up-move versus a down-move. That instinct is understandable, but it’s not quite how the no-arbitrage version of this model works, and getting this right matters enormously for the exam.

The binomial model prices the option by constructing a replicating portfolio — a combination of the underlying stock and risk-free borrowing or lending that produces exactly the same payoff as the option in both the up-state and the down-state. If you can build a portfolio of stock and cash that perfectly mimics the option’s payoff no matter which branch occurs, then today’s option value must equal today’s cost of that replicating portfolio. Otherwise, an arbitrage opportunity would exist — you could buy the cheaper of the two and sell the more expensive one, locking in riskless profit, which can’t persist in an efficient market.

Working through the numbers: you need to find a hedge ratio, h, representing how many shares of stock to hold per option, such that the portfolio payoff matches in both states.

In the up-state: h × ₹120 − (loan repayment) = ₹20 (matching the call payoff)

In the down-state: h × ₹90 − (loan repayment) = ₹0 (matching the call payoff)

Solving these two equations together, the hedge ratio h comes out to (₹20 − ₹0) / (₹120 − ₹90) = ₹20 / ₹30 = 0.667. This means holding roughly two-thirds of a share of stock per call option replicates the option’s payoff exactly, in both possible future states.

From there, you can work out exactly how much to borrow today so that the replicating portfolio’s cost matches the option’s fair value, and discount everything back to today using the risk-free rate. The mechanical details of that final discounting step vary slightly depending on whether you’re using the replicating portfolio approach directly or the risk-neutral probability shortcut — but the core backward-looking logic stays identical either way: known terminal payoffs get translated, one step at a time, into a value today.

The Risk-Neutral Shortcut

In practice, most of the curriculum moves toward a faster version of this same logic called risk-neutral valuation, since constructing an explicit replicating portfolio at every single node of a multi-period tree gets tedious fast.

The risk-neutral approach calculates a risk-neutral probability, π, of an up-move — a probability that’s mathematically derived to make the expected stock return equal the risk-free rate, not the actual real-world probability of the stock going up. Using that derived probability, the option’s value at any node becomes the probability-weighted average of its two possible next-period values, discounted back one period at the risk-free rate.

It’s worth being honest about why this shortcut is allowed: it isn’t a different model. It’s mathematically equivalent to the replicating portfolio approach — the risk-neutral probability is constructed precisely so that pricing this way produces the identical answer you’d get from building the replicating portfolio directly. The “shortcut” just hides the replication mechanics inside a probability-weighted formula, which is faster to apply once you trust that the underlying logic holds.

Either way — replicating portfolio or risk-neutral probability — backward induction is doing the same job: starting from certain terminal payoffs and working one period backward at a time.

Extending to Multiple Periods: Where Backward Induction Really Shows Its Value

A single-period tree barely needs the word “induction” at all — there’s only one step backward to take. The technique earns its name properly once you move to a multi-period tree, say a two-period binomial model used to price an American-style option.

Picture a tree with today at time 0, an intermediate point at time 1, and expiration at time 2. From today’s node, the stock can move up or down to reach time 1. From each of those time-1 nodes, the stock can again move up or down, landing at one of three possible nodes at time 2 (since an up-down path and a down-up path land at the same middle value, assuming a recombining tree).

Backward induction here means: first calculate the option’s value at every node at time 2 — trivial, since these are just terminal payoffs. Then move one step back to time 1, and at each of the two time-1 nodes, use the already-known time-2 values (the ones directly reachable from that node) to calculate that node’s value, exactly the way you did in the single-period example. Finally, move back one more step to time 0, using the two now-known time-1 values to calculate today’s option value.

Where This Gets Genuinely Important: American Options

Here’s the part of backward induction that goes beyond just “discounting a tree neatly,” and it’s the part most likely to actually get tested at a meaningful level: early exercise.

For a European option, backward induction is purely about calculating the discounted, probability-weighted value at each node — straightforward repetition of the same formula, moving backward through the tree. But for an American option, which can be exercised at any point before expiration, something extra has to happen at every single intermediate node: you have to compare the calculated “hold” value — what the option is worth if you keep it un-exercised and let it continue through the tree — against the “exercise now” value — the immediate intrinsic value if exercised at that exact node — and take the larger of the two.

This comparison is precisely why backward induction is the only sensible way to price American options in a binomial framework. You genuinely cannot know whether early exercise is optimal at an intermediate node without first knowing the option’s continuation value from that node forward — and the continuation value itself depends on values further along the tree that haven’t been calculated yet, until you’ve worked backward to them. Try to value the tree forward, from today outward, and you’d be stuck not knowing what “continuing to hold” is actually worth at any point, because you wouldn’t yet have valued the later nodes that determine it.

A Worked Two-Period Illustration

Let’s make this concrete with numbers, using a hypothetical American put option, since puts are where early exercise considerations show up most often in this kind of example.

Suppose a stock starts at ₹100. Each period, it can move up by a factor of 1.2 or down by a factor of 0.85. You’re pricing a 2-period American put with a strike price of ₹105.

At time 2, the three possible stock prices are: up-up = ₹100 × 1.2 × 1.2 = ₹144, up-down (same as down-up) = ₹100 × 1.2 × 0.85 = ₹102, and down-down = ₹100 × 0.85 × 0.85 = ₹72.25.

Put payoffs at time 2 (since a put is worth max(0, Strike − Stock)): at ₹144, the put is worth ₹0 (deep out of the money). At ₹102, the put is worth max(0, ₹105 − ₹102) = ₹3. At ₹72.25, the put is worth max(0, ₹105 − ₹72.25) = ₹32.75.

Now step back to time 1. At the time-1 up-node (stock at ₹120), the relevant time-2 values reachable from here are ₹0 (if it goes up again to ₹144) and ₹3 (if it goes down to ₹102). Using the risk-neutral probability-weighted approach and discounting back one period gives you a “hold” value for this node — let’s say that calculation produces approximately ₹1.40 once probability-weighted and discounted (the exact figure depends on the risk-free rate assumed, which isn’t the focus of this illustration). Compare that to the “exercise now” value at this node: max(0, ₹105 − ₹120) = ₹0, since the stock is above the strike and exercising the put immediately would be worthless. Holding clearly beats exercising here, so the time-1 up-node value is the ₹1.40 hold value.

At the time-1 down-node (stock at ₹85), the relevant time-2 values reachable are ₹3 (up to ₹102) and ₹32.75 (down to ₹72.25). The probability-weighted, discounted hold value here comes out considerably higher — say, roughly ₹16.50 for illustration. Compare that against the exercise-now value at this node: max(0, ₹105 − ₹85) = ₹20. Here, exercising immediately is worth more than holding. This is exactly the situation that makes American options different from European ones — the model says exercise early, right here at this node, rather than waiting. So the time-1 down-node value becomes ₹20, the early-exercise value, not the smaller hold value.

Step back once more to time 0, using these two time-1 values — ₹1.40 from the up-node and ₹20 from the down-node — probability-weighted and discounted one final period, to arrive at today’s option value.

Notice what just happened structurally: you couldn’t have known, sitting at time 0, whether early exercise would matter anywhere in this tree without first working through time 2 and time 1. Backward induction isn’t optional bookkeeping here — it’s the only order of operations that actually lets you capture the early-exercise decision correctly.

Why Forward Valuation Doesn’t Work

It’s worth being explicit about why you can’t just value this tree moving forward from today, since understanding the failure mode helps cement why backward induction is structurally necessary rather than just a convention.

Valuing forward would require knowing, at time 0, what the option will be worth at time 1 — but that time-1 value itself depends on knowing what happens at time 2, and possibly on an early-exercise decision that depends on comparing time-2-derived continuation value against immediate exercise value. There’s no way to compute that comparison without already having the time-2 values in hand. The dependency runs strictly backward: every node’s value depends on values further along the tree, never on values before it. Backward induction simply respects that dependency structure, solving things in the only order that’s actually logically available.

Exam Perspective: What to Carry Into the Test

A handful of points are worth locking in firmly. Backward induction works because terminal option payoffs are known with certainty, allowing you to start there and work backward one period at a time, rather than needing to forecast stock price direction. The risk-neutral probability approach is mathematically equivalent to building an explicit replicating portfolio at each node — it’s a computational shortcut, not a different pricing philosophy. For European options, backward induction is purely mechanical discounting through the tree. For American options, every intermediate node requires comparing the calculated hold value against the immediate exercise value, taking the larger of the two — and this early-exercise check is precisely why backward induction is structurally required rather than just procedurally convenient.

Final Thoughts

Backward induction is one of those techniques that initially feels like a clever trick and later feels like the only sensible way the problem could ever have been solved. You can’t know what something is worth today without knowing what it leads to tomorrow — and in a binomial tree, “tomorrow” is only fully knowable once you’ve already solved “the day after that.”

Start where the answer is certain. Work backward, one period at a time, letting each known value inform the value just before it. By the time you’re back at today, you haven’t guessed anything about the future — you’ve simply let the future’s certainty flow backward into the present.