BackstedgeBlog
Start for free
← All articles

No xG Betting Strategy: Five Season Backtest Using Goals Averages

No xG Betting Strategy: Five Season Backtest Using Goals Averages

Isometric goals averages backtest title card

In this article, an xg betting strategy means a rule-based approach built on real goals scored and conceded, not expected goals models. The fastest path is to define a goals-averages rule, then backtest it against historical closing odds using a tool like Backstedge, which runs exactly that kind of test without spreadsheets or code.

TL;DR:

  • Strategies should rely on actual goals scored and conceded averages from recent matches rather than estimated expected goals, ensuring factual inputs.
  • The backtest requires a large sample size, ideally spanning multiple seasons with at least several hundred qualifying bets, to reliably measure profitability.
  • Using flatten one-unit stakes is recommended for practical, transparent bankroll management, with bet size tied to the worst historical drawdown to prevent overexposure.
  • Backstedge automates rule creation and testing against real historical odds, providing stable season-by-season metrics to assess strategy consistency.
  • Focus on real closing odds rather than opening prices, as they better reflect market consensus and improve the accuracy of identifying genuine mispricings.
Backstedge
backstedge.com
Turn Betting Ideas Into Tested Strategies
Build rules from real match data, backtest them across historical odds, and track their stability without coding or spreadsheets.
Test your strategy

Table of Contents

How to design a rule-based strategy using goals averages

Expected goals (xG) estimates the quality of chances a team creates, weighting shots by location, angle and situation to guess how many goals they “should” score. It is a modeling choice, not a measurement, and it depends on data providers who score chances differently. This guide skips that entirely. Backstedge’s rules run on real goals scored and conceded averages over a team’s last ten matches, which means every input is a fact from the scoresheet, not an estimate.

Turning that into something you can test takes four steps.

  1. Pick your league and sample window first. Stick to one league or a small group of similar leagues, and favor several contiguous seasons over a single short stretch. A double Poisson model study of Euro 2020 predictions found that model performance is sensitive to where the data window starts and ends, so test more than one start date before locking in a rule.
  2. Compute rolling goals averages. A common baseline is each team’s goals-for and goals-against average over its last 10 matches. For example, a rule might require both teams to show a last-10 goals-for average of 1.6 or higher before a match qualifies.
  3. Layer in a closing-odds filter. Set a minimum decimal odds threshold and an implied probability band so the rule avoids extreme longshots and heavily backed favorites alike. This keeps the strategy anchored to what the market already believes, rather than betting blind against it.
  4. Write the rule so a machine could run it. Specify the exact fields (goals for, goals against, match count), the look back window (last 10 matches, not “recent form”), and the date cutoff separating training data from live testing.

The result should read like a specification, not an idea: “when both teams’ last-10 goals-for average is at least 1.6 and closing odds on over 2.5 goals are between 1.70 and 2.20, place one unit.”

Assemble the backtest: required data, odds alignment and simulation rules

A backtest is only honest if the data lines up the way a real bet would have. Before running any simulation, gather full-time goals, match dates, home or away status and closing odds for every match in your sample, then check that the odds attached to each row were actually available before kickoff, not adjusted afterward.

  • Pull full-time goals, date, venue side and closing odds into one table, matched row by row.
  • Order matches chronologically and make sure no bet uses information from a match that happened after it.
  • Use a walk-forward split: build or calibrate the rule on an earlier window, then test it on a later one it has never seen.
  • Pick a stake model up front. Flat one-unit staking is the simplest baseline and the easiest to audit.
  • Document what happens to postponed matches, missing odds or abandoned games rather than quietly dropping them.
  • Resist tuning thresholds match by match until the backtest looks good. A review of replication issues in sports betting backtests found that profitability in published results often disappears once outlier bets or data errors are removed, which is a strong argument for simple, interpretable thresholds over heavily fitted ones.

Reproducible workflows built by other betters follow a similar shape: freeze the rule on a training window, test it out of sample on the following season, then roll the window forward one season at a time. That structure, described in a public walk-forward backtest of EPL prediction models, prevents a rule from accidentally learning from the future.

Which metrics to report and what levels of evidence matter

A single ROI number tells you almost nothing on its own. A credible backtest reports several figures together, and it reports them per season, not just as one blended total.

  • ROI: net profit divided by total staked, the headline figure but never the only one.
  • Gross profit: the raw unit return, useful for sanity-checking ROI against bet count.
  • Worst drawdown: the deepest losing streak the strategy would have lived through.
  • Number of bets: how many qualifying matches the rule actually found across the sample.
  • Win rate: the share of bets that settled as winners, read alongside odds level, not alone.
  • Season-by-season breakdown: whether the edge held up every year or came from one outlier season.

A large-sample study analyzing 479,440 games across 818 leagues from 2005 to 2015 found that a strategy exploiting mispriced closing odds returned a positive ROI in both simulation and paper trading, which is the kind of sample size that makes a result worth trusting. Small samples cut the other way: a rule that fires on 40 matches can post an eye-catching ROI purely from a handful of results going one way. Favor rules that generate at least a few hundred qualifying bets across multiple seasons, and treat a strong result on a thin sample as a hypothesis, not a conclusion.

Bankroll and staking: practical rules tied to observed drawdowns

Flat one-unit staking is the right default for most rule-based strategies because it keeps every result comparable and makes a backtest easy to audit. Fractional Kelly staking can outperform flat stakes in theory, but it depends on an accurate estimate of your edge, and a goals-averages rule with a noisy hit rate will feed Kelly bad inputs, which tends to produce oversized bets exactly when you are least sure of your edge.

Diagram of drawdown-based staking rules

A more practical approach ties unit size to the worst drawdown observed in the backtest. If a strategy’s worst historical losing run cost 12 units, size each bet so that stretch would not threaten your bankroll or your composure. Add operational limits on top: a cap on bets per day or per league, a maximum number of concurrent open positions, and a stop-loss rule that pauses the strategy after a losing streak beyond what the backtest ever showed.

Pro Tip: Size your unit from the backtest’s worst drawdown, not from its average return, since the drawdown is what you actually have to survive.

What the data says: Backstedge’s over 2.5 high-scoring teams backtest

Backstedge does not model xG. Its “over 2.5 when both teams score freely” strategy is a close real-world proxy for the idea behind an xG-flavored rule, built entirely on goals-averages filters and real closing odds rather than any expected-goals estimate. The over 2.5 high-scoring teams strategy page publishes the full backtest across five seasons (2021/22 to 2025/26) and five major football leagues, using flat one-unit stakes against real closing odds.

ROI, worst drawdown, number of bets and win rate are reported on the strategy page itself, broken down season by season across the five-year sample.

Those figures reflect what already happened in five completed seasons under real market prices, not a forecast of what will happen next season. Treat them as evidence that the rule type has historical merit, and re-run the backtest yourself before trusting it with live stakes.

How to test your own rules now

Three steps get you from idea to result: define a specific rule (goals averages plus an odds band), choose a sample window of several contiguous seasons, and run a flat one-unit backtest against real closing odds. Backstedge builds all three into one no-code workflow, so the rule you write in step one is the exact rule the backtest runs in step three, with automated tracking that flags future matches meeting your criteria as they come up.

Backstedge

That combination matters most for stability analysis: a rule that looks profitable overall but falls apart in one season out of five is a different proposition than one that holds steady, and Backstedge’s season-by-season view is built to surface that difference before you stake real money on it.

Selected research and Backstedge pages used in this article

Everything in this guide, the rolling goals averages, the closing-odds filters, the walk-forward backtest, the season-by-season metrics, is exactly what Backstedge is built to run without spreadsheets or code. You set the thresholds visually, Backstedge checks them against years of historical results and real closing odds, and it flags upcoming matches that qualify under your rule automatically.

Backstedge as the recommended solution for your strategy testing — overview diagram

That fits a reader who has just worked through the mechanics above and wants to see them applied without rebuilding a database from scratch. The Free, Pro and Advanced plans give you a place to build the rule you just designed, backtest it over multiple seasons, and check its stability before you ever place a live bet on it. Start with the free plan, build your goals-averages rule, and see what five seasons of real odds say about it.

Sources

FAQ

Is an xG betting strategy the same as expected goals modeling?

Not in the sense used here. This guide treats an xg betting strategy as a rule built on real goals scored and conceded averages, explicitly not on expected goals estimates, because goals averages come straight from results rather than a modeling choice.

How many seasons of data should a backtest use?

There is no fixed number, but favor several contiguous seasons over one short stretch, since research on goals-average models found that results are sensitive to where the sample window starts and ends. Backstedge’s own published strategy backtests often run over multiple seasons for this reason.

What is a reasonable sample size before trusting a strategy’s ROI?

Favor rules that generate several hundred qualifying bets across multiple seasons rather than a strong result on a few dozen matches. A large-sample study covering 479,440 games across ten years is the scale that makes a closing-odds edge credible, and season-by-season consistency matters as much as the headline number.

Why use closing odds instead of opening odds in a backtest?

Closing odds reflect the market’s final view after news, injuries and money have shifted prices, which makes them a stronger baseline for judging whether your rule found a genuine mispricing. Backstedge’s published backtests, including the over 2.5 high-scoring teams strategy, are built entirely on real closing odds for this reason.

Can Backstedge build strategies for sports other than football?

Backstedge’s published betting strategies, including the goals-averages backtests referenced in this guide, cover football (soccer) only. If you bet on other sports, the rule-building approach described here does not currently apply on the platform.