All payments made in the preview are in test mode. Read more
All case studies
Case study · NRL · Backtest

Kelly staking vs flat staking — 2024 NRL backtest

Same model, same 384 picks across the 2024 NRL regular season. Two staking plans: flat 1 unit, and fractional Kelly at one-quarter. The bankroll curves diverge in a way most punters underestimate.

7 min readNRLKellyBacktestBankroll

Flat staking is comfortable. Every bet is the same size, the maths is trivial, and a losing streak feels survivable. Kelly is uncomfortable — bets vary, big edges sting on losses, and the formula tells you to do things your gut won't. The 2024 NRL backtest is a fair fight between the two, run on the same picks the Edgewise NRL ensemble actually produced.

The setup

Same 384 picks across H2H and Line markets, edge filter ≥ 3 pp, starting bankroll $10,000. Both plans bet only when the scanner flagged an edge — no martingale, no doubling-up, no skipping picks based on feel.

Picks
384
Win rate
54.2%
Average edge
4.6 pp
Average odds
$1.94

Result — final bankrolls

Flat 1u (1%)
+18.4%
$11,840 final
¼ Kelly
+34.7%
$13,470 final
Max DD flat
−6.1%
Max DD Kelly
−11.8%

Kelly produced nearly 2× the return — at almost 2× the drawdown. That trade-off is not free, and it's the central question of any staking plan.

What's actually happening

Flat staking

A 1-unit flat bet is, mathematically, a constant fraction of the starting bankroll. As the bankroll grows, your effective stake-as-percent shrinks, and you under-bet your edge. The flat plan ended the season risking roughly 0.85% per pick, even though it could have safely risked more.

¼ Kelly

Kelly recalculates after every settled bet, so the stake-as-percent stays roughly constant in compound terms. When you're winning, you press; when you're losing, you reduce. The result is exponential rather than linear growth — at the cost of deeper drawdowns when a cold patch hits.

Why not full Kelly

Full Kelly is theoretically optimal when your true probabilities are known with certainty. They're not. Sports models have estimation error, and full Kelly punishes that error brutally — a 5 pp model overestimate at full Kelly can wipe out 20–30% of a bankroll in a single bad week. Fractional Kelly (¼ or ⅓) keeps most of the geometric-growth benefit while dampening the variance you actually face.

The pitfall flat hides

Flat staking looks safer because the numbers feel steady, but it's actively wasting edge. Over 384 picks with a real 4.6 pp average edge, the flat plan left roughly $1,630 on the table — purely from under-betting. That gap compounds. Across three seasons, the same picks under flat vs ¼ Kelly produce a ~60% bankroll gap.

How to run this yourself

Open Backtest → Sport = NRL → choose any NRL model → Strategy Comparison tab. Set strategy A to flat 1% and strategy B to ¼ Kelly with the same edge filter. The engine will produce the same side-by-side curves you see in this post on your own bankroll size and risk preferences.

EdgewiseedgewiseAustralian sports intelligence
Important · educational analytics only · available Australia-wide

Edgewise provides quantitative sports analytics for research and education to users across all Australian states and territories. Nothing on this platform is financial, investment, legal, tax or betting advice and no profits, returns or outcomes are guaranteed. Models are statistical estimates and may be wrong. Edgewise is not a bookmaker, does not accept wagers, does not offer or promote inducements, and is not a regulated wagering service provider under the Interactive Gambling Act 2001 (Cth).

You win some. You lose more. Gambling carries financial risk and can be addictive. You must be 18 or older to use this service anywhere in Australia. Bet only with operators licensed in an Australian state or territory and only within your means. Free, confidential 24/7 support: Gambling Help Online — 1800 858 858 · gamblinghelponline.org.au. National self-exclusion register: betstop.gov.au. Personal information is handled in accordance with the Privacy Act 1988 (Cth) and the Australian Privacy Principles.

© 2026 Edgewise · Sydney, Australia