Is Kelly the same for tennis and football?

The math is identical, but tennis has higher per-bet variance because matches resolve in 1-3 sets with strong serve-hold dynamics. That argues for a smaller Kelly fraction in tennis (0.10-0.25) vs longer-tail sports.

What if I don't know my real edge?

Then don't use Kelly — use flat stakes. Kelly's only useful when you have a calibrated probability estimate; without one, fractional Kelly degenerates to over-betting on noise.

Why does TIPERO use 0.15 Kelly specifically?

It's roughly midway between quarter-Kelly (conservative) and tenth-Kelly (very conservative). The 0.15 setting matched empirical bankroll smoothness in TIPERO's backtests across 4,000+ historical bets.

The Kelly Criterion for Tennis Bettors: Practical Guide

Kelly tells you how much to bet given an edge. Done right it maximises long-run growth. Done wrong it's the fastest way to blow up a bankroll. Here's the full tennis-specific picture.

The basic Kelly formula

For a single binary bet:

f* = (p × b − q) / b

where:
  p = your estimated probability of winning
  q = 1 − p
  b = decimal odds − 1  (i.e. net payout per unit staked)
  f* = fraction of bankroll to bet

If your model says Sinner wins with p = 0.60 at odds of 1.85 (b = 0.85):

f* = (0.60 × 0.85 − 0.40) / 0.85
   = (0.51 − 0.40) / 0.85
   = 0.129  →  bet 12.9% of bankroll

Why nobody actually bets full Kelly

Full Kelly is mathematically optimal if your probability estimate is exact. In reality your p is a noisy estimate. Even a small overestimate of p produces wildly oversized bets and crushing drawdowns. The standard solution is fractional Kelly:

stake = f* × kelly_fraction

  Common settings:
  - 0.5 Kelly (half-Kelly)   — used by many syndicates
  - 0.25 Kelly (quarter)     — what TIPERO ships by default
  - 0.10 Kelly               — extremely conservative

Quarter-Kelly cuts max drawdown by ~75% while keeping ~94% of long-run growth — the best risk-adjusted point on the curve for most bettors.

Why tennis is hard for Kelly

Probability uncertainty — even a great model is ±3-5% on any single match.
Correlation — multiple picks on the same day are correlated through fatigue, weather, draw position.
Tail risk — three-set matches with tiebreaks have heavy tails that Kelly's normal-ish assumptions ignore.
Liquidity — Kelly assumes you can place any size; in practice limits cap big bets.

Practical adjustments for real-world tennis betting

Use fractional Kelly — TIPERO uses 0.15 (between 0.10 and 0.25) as default.
Cap stakes per bet — even if Kelly says 8% of bankroll, cap at 1.5u.
Cap stakes per day — if 6 picks each suggest 0.5u, scale back when correlated.
Drop Kelly when EV is marginal — below ~3% EV, fixed flat-stakes is more robust than Kelly noise.

What TIPERO actually does

The stake_sizing_apply.py step in the daily pipeline computes Kelly per pick using the model's p and the listed odds, applies a 0.15 Kelly fraction, then caps stakes at 0.25u min and 1.5u max. Tier-specific caps further constrain ULTRA picks (long-shot odds) to 0.30u so a single bad ULTRA can't destroy a week.

Bottom line

Kelly is a fantastic theoretical tool but a dangerous practical one. Fractional Kelly with hard caps is what real syndicates use. TIPERO ships pre-sized stakes so you don't have to do the math.

Get pre-sized Kelly picks →