Betfair Historical Data: Where to Get It and How to Use It

This page contains affiliate links — if you open an account through them we may earn a commission at no cost to you. It never changes our verdict.

This page is a sub of our data and analytics pillar, and it tackles the first practical wall every aspiring quant hits: getting the actual prices. You can theorise about a strategy all day, but until you've replayed it against real recorded market movement, you don't know whether it has an edge or whether you've just told yourself a nice story. Betfair, to its credit, sells the real recorded data — the catch is that it doesn't come in a form anyone can just open in Excel and read.

Why you need historical data

Backtesting is the difference between a hunch and an evidenced edge. A claim like “well-backed favourites tend to drift slightly in the final ten minutes before the off” is testable — but only if you have the prices for thousands of races at minute-by-minute resolution. Live observation gives you a few dozen data points and a strong dose of confirmation bias; historical data gives you thousands and the chance to be proven wrong, which is the whole point. Good data lets you measure the size of an edge, the variance around it, how often it fails, and what commission and realistic fills do to it. Without it you're trading on vibes, and the exchange is unusually efficient at removing money from people trading on vibes. The corollary is that bad or incomplete data is almost worse than none, because it gives you false confidence — which is why where you get the data, and how complete it is, matters enormously.

The official Betfair Historical Data service

The authoritative source is Betfair's own Historical Data service, hosted at historicdata.betfair.com and accessible with your Betfair account. It contains recorded exchange data from 2016 onwards across the major sports — horse racing, football, tennis, greyhounds and more — organised by sport, market type and date. Crucially, this is a recording of the actual data feed: the same market-change messages the live API streams, captured and stored. That means it reflects the real price ladder and, in the higher tiers, the real traded volumes, rather than some summarised end-of-event snapshot. You buy it per month of data, per sport, and download compressed archives. Because it's the genuine feed replayed, it's the only source that lets you reconstruct exactly what a trader would have seen tick by tick — which is precisely what a credible backtest requires.

BASIC, ADVANCED and PRO: what each gives you

The service comes in three tiers and the differences decide what you can actually test. BASIC is free and limited — it gives you market definitions and final results with a coarse view of prices, enough to learn the file format and do simple outcome studies, but not enough to model a trading strategy that depends on the moving ladder. ADVANCED adds the price data you need: best available back and lay prices over time, so you can see how the market moved and model entries and exits at prices that genuinely existed. PRO goes further again, adding full ladder depth and traded volume — every price that traded and how much — which is what you want if your strategy cares about liquidity, queue position, or how much you could realistically have got matched. As a rule: BASIC to learn the format, ADVANCED for most trading backtests, PRO if liquidity and fill realism are central to your edge. Pricing is monthly per sport, so a focused study of one sport over a defined period is far cheaper than hoovering up everything.

Free and third-party sources

Before you pay, know the free routes and their limits. Betfair publishes sample datasets for each tier — a handful of markets you can download for nothing, which is the right way to learn the parsing before you spend anything. The BASIC tier itself is free and fine for results-based studies. Beyond Betfair, a few third-party datasets and academic archives circulate, and there are community projects that share parsed samples; quality and completeness vary wildly, and licensing is often murky, so treat them as a way to experiment rather than a foundation for real money. The other “free” option is to record your own live data going forward via the Betfair API — you set up a stream listener, capture markets you care about, and build your own archive over weeks and months. That costs you time rather than money and gives you exactly the markets you want, but obviously can't give you the past. For anything historical and serious, the official ADVANCED or PRO feed remains the only complete answer.

What the files actually look like

This is where most people stall, so let's be concrete. A Betfair historical file is not a CSV of prices. It's a compressed archive (typically a .tar of .bz2 files) and inside each is a stream of JSON market-change messages — one line per change. Each message says, in effect, “at this timestamp, on this market, the back price on selection X is now this, the available amount is now that, the last traded price moved to this.” In other words it's a diff stream: it tells you what changed, not the full state each time. To know the complete ladder at any given second, you have to start from the market definition and apply every change message in order, maintaining the current state as you go. That's a different mental model from a spreadsheet, and it's the single biggest reason people give up — they open the file expecting rows of prices and find a wall of incremental JSON.

Turning raw files into usable data

The workflow is always the same shape. Decompress the archive, then read each market file line by line, parsing the JSON. Maintain state: hold the current best back, best lay and last traded price for each selection, updating them as each change message arrives, and stamp each state with its timestamp. Sample or snapshot to the resolution you need — for most trading studies a snapshot every second, or even every five seconds, around the period you care about is plenty, and far easier to work with than every micro-change. Then flatten the result into a tidy table: one row per selection per timestamp, with the columns you'll analyse. Betfair publishes example parsers, and the community has built libraries — the Python data analysis tutorial walks through doing exactly this with pandas, and the API with Python guide covers the same message format from the live side. Once you have that tidy table, the hard part is done and the actual strategy logic is comparatively easy.

Pitfalls that ruin a backtest

A few mistakes turn a backtest into self-deception. Assuming perfect fills is the worst: in the data you can “match” at any price you see, but live you have to queue, and in a thin market you might not get matched at all — always model realistic liquidity, which is why PRO's traded-volume data matters for serious work. Ignoring commission flatters every result; a thin edge often vanishes once you subtract 2–5%. Over-fitting is the seductive one: if you tweak entry and exit rules until the past looks perfect, you've fitted noise, not signal, and it won't survive live — keep rules simple and test them on a period you didn't tune them on. Look-ahead bias sneaks in when your logic accidentally uses information that wasn't available at the moment of the trade. And survivorship and sample-size problems mean a month of data and forty trades prove very little. Treat a backtest as evidence to be argued with, not a green light — the market that generated the data is also the market that will happily punish overconfidence.

How much data do you actually need?

A mistake almost everyone makes is buying too much data and then drowning in it. The instinct is “more is better — give me every sport, every month, going back to 2016”, and the result is hundreds of gigabytes of compressed streams you'll never parse and a bill you didn't need to pay. The right question is narrower: what is the smallest dataset that can answer my specific question? If you're testing a pre-race drift signal on UK win markets, you need UK racing win markets for a representative period — and a couple of months across different parts of the season is usually enough to see whether an effect is real and roughly how big, before you spend more confirming it. Buying one focused sport for a defined window is cheap; hoovering up everything is expensive and slows every parse you run. The other dimension is resolution: you rarely need every micro-tick. For most trading studies, a one-second snapshot is finer than you'll actually use, and for slower signals a five-second snapshot makes files dramatically smaller and analysis faster with no real loss. There's also a sensible workflow: prototype on free samples to get your parser and logic working, buy a small focused slice to get a first read on the edge, and only scale up the date range or sports if the early read justifies the spend. That staged approach keeps costs down and, more importantly, keeps you from confusing “I have lots of data” with “I've learned something” — they're not the same, and a tightly scoped study on two months of one market will teach you more than a terabyte you never finish parsing.

The verdict

If you're serious about testing a Betfair strategy, the official Historical Data service is where to start — BASIC and the free samples to learn the format, ADVANCED for most trading backtests, PRO when liquidity and fills are central. The data is genuine and complete from 2016, which is more than most markets give you. The real work isn't buying it; it's parsing the stream into something tidy and then resisting the urge to over-fit. Build your own live archive via the API alongside it, lean on the Python workflow, and anchor everything in the data and analytics pillar. Get the data right and you can finally trade on evidence instead of memory — just remember that a backtest measures the past, and the market only ever pays you in the present.

FAQ

Related reading

This is a sub of our Betfair data and analytics pillar. Pair it with the best data sources and APIs roundup and the hands-on Python data analysis tutorial. To collect your own live data instead of buying it, see the Betfair API guide, the API with Python walkthrough and our guide to building Betfair bots. Then apply what you learn to pre-race and swing trading.