The question
"There's twice as much money waiting to back as to lay. Does Betfair weight of money tell me which way the price will go next?"
Every trader stares at the ladder and feels the pull of the big numbers. This lesson turns that feeling into something you can measure, and shows how to test whether it really predicts anything.
The idea in one sentence
Weight of money compares the unmatched money waiting to back with the unmatched money waiting to lay, and it's only useful if a proper test shows it predicts the next move by enough to pay after commission.
The picture
The Betfair ladder has two sides. Each shows money that someone has offered but no one has taken yet:
- The back column (blue) shows prices you can back at right now. The money there comes from layers waiting to be matched.
- The lay column (pink) shows prices you can lay at. The money there comes from backers waiting to be matched.
Here is an illustrative snapshot of Over 2.5 goals:
| Backers waiting (lay column) | Price | Layers waiting (back column) |
|---|---|---|
| £3,800 | 1.93 | |
| £2,800 | 1.92 | |
| £3,400 | 1.91 | |
| 1.90 | £1,200 | |
| 1.89 | £2,100 | |
| 1.88 | £1,700 |
There's £10,000 waiting to back and £5,000 waiting to lay. The story traders tell is simple: more people want to back, so the price will shorten. The real question is whether that story holds up often enough to make money.
Three things weaken it:
- Money in the queue is free until it's matched. Some of it is placed to create an impression and cancelled before anyone reaches it.
- Big orders are split up. The real intention of a large, well-informed player often never shows on the ladder at all.
- Where you stand in the queue matters. If you join the back of a queue, you tend to get matched just as the price moves through your level, which is often the moment it's about to go against you (queue position).
Worked Betfair example
You want to test one rule: "When the imbalance across the top three prices is above +0.3, the price will shorten in the next 60 seconds." You record snapshots across many pre-match Over 2.5 markets. (Illustrative figures.)
- Measure the imbalance. In the snapshot above: (£10,000 − £5,000) ÷ (£10,000 + £5,000) = +0.33. That's above the +0.3 trigger.
- Collect the cases. You find 400 moments where the trigger fired and the price moved within 60 seconds. It shortened in 228 of them (57%).
- Test it. If weight of money meant nothing, you'd expect 200 shortenings, give or take √(400 × 0.5 × 0.5) = 10. You're 2.8 standard errors above that. The exact chance of 228 or more from 400 coin tosses is 0.003.
- Put a price on it. Suppose you back £100 at 1.91 when the trigger fires and green up one tick either way. Right: lay at 1.90, locking £100 × (1.91 ÷ 1.90 − 1) = £0.53, or £0.52 after 2% commission. Wrong: lay at 1.92, locking −£0.52.
- Expected value per trade. 0.57 × £0.516 − 0.43 × £0.521 = +£0.07 per £100 traded. Break-even is a hit rate of 50.2%, so there's room, but not much.
- The honest checks. This one rule was planned in advance, so the p-value means what it says. If you'd tried 30 thresholds and time windows, it wouldn't (Lesson 7.4). And the 57% must be measured at the prices you'd actually have been matched at, including the times your order in the queue didn't get filled.
Verdict: a weight-of-money signal can carry a little information, but the edge per trade is thin and easily wiped out by testing too many variations. Treat it as a hypothesis to prove on fresh data, never as a signal to follow on sight.
Matched money says more than unmatched money
Betfair also shows how much has been traded at each price. That's money people actually committed. A common summary is the volume-weighted average price (VWAP). If £4,000 has traded at 1.88, £10,000 at 1.90 and £6,000 at 1.92, the VWAP is (1.88 × 4,000 + 1.90 × 10,000 + 1.92 × 6,000) ÷ 20,000 = 1.902. A price sitting well away from its VWAP is one thing worth testing; see weight of money and VWAP.
The formula
Order-book imbalance
- B is the unmatched money waiting to back (shown in the lay column) across the prices you choose, say the best three.
- L is the unmatched money waiting to lay (shown in the back column) across the same depth.
In plain English: I runs from −1 to +1. Positive means more backers are waiting, negative means more layers. Zero means balanced.
Testing the hit rate
- k is the number of times the price moved the way the imbalance pointed.
- n is the number of times the trigger fired and the price moved.
In plain English: how far your hit count sits above a coin toss, in standard errors. Above about 3, with a rule you fixed in advance, is worth a closer look.
Volume-weighted average price
- P_j is each price that has traded, and V_j is the amount matched at it.
In plain English: the average price the market has actually traded at, with busier prices counting for more.
Try it
Pen and paper: backers waiting across the top three prices total £900 + £600, and layers waiting total £2,000 + £1,500. What's the imbalance, and which way does the naive story say the price will go?
Answer: (1,500 − 3,500) ÷ 5,000 = −0.4. More money waiting to lay, so the naive story says the price drifts. Now you'd have to test whether that's true.
Common mistakes
- Believing the biggest number on the ladder. Money that can be pulled in a click isn't a commitment. Test what the queue predicts; don't assume it.
- Testing on the moments you remember. The times weight of money "called it" stick in the mind. Record every trigger automatically, including the ones that failed.
- Trying lots of thresholds. +0.2, +0.3, +0.4, top one price, top five prices: each is another test. Count them.
- Ignoring which orders get filled. A signal measured on prices you wouldn't have been matched at is a backtest that lies (Lesson 7.5).
- Forgetting that the edge per trade is tiny. A few pence per £100 means commission and a slightly lower hit rate can turn it negative. Know the break-even before you trade.
The market is full of people reading the same ladder as you: why good models still lose money.