The Kelly Criterion was developed by John L. Kelly Jr. at Bell Labs in 1956, originally as a framework for signal transmission noise. Edward Thorp subsequently adapted it for blackjack card counting and later for financial markets. Today it is the foundational bet sizing formula across professional gambling, trading, and investment management. It answers a simple but critical question: given known probabilities and payouts, how much of your bankroll should you risk on a single bet?
For 4D lottery markets — which are negative expected value games — the Kelly output is also simple but uncomfortable. This guide presents the full calculation, explains what it means for lottery participation, and proposes a practically useful adaptation of the Kelly framework for disciplined 4D bankroll management.
The Kelly Formula
For a binary bet (win or lose), the Kelly Criterion gives the optimal fraction of bankroll to bet as:
Where:
f* = fraction of bankroll to bet
b = net odds received per unit staked (decimal odds - 1)
p = probability of winning
q = probability of losing = 1 - p
For a bet with a positive expected value, f* is positive — Kelly tells you to bet a positive fraction of your bankroll. For a bet with a zero expected value, f* is zero — Kelly tells you not to bet. For a bet with a negative expected value, f* is negative — Kelly tells you to take the other side of the bet if possible, or not to bet at all.
Applying Kelly to Singapore 4D: The Exact Output
Take the Singapore 4D ordinary Big bet for a single number. The probability of any win (any prize tier) is approximately 22/10,000 = 0.0022. To apply Kelly to the "win/lose" binary, we simplify to the most likely scenario — no prize (99.78% of draws):
For the Starter Prize tier specifically (10 winning numbers out of 10,000, net payout $227.50 per $1 staked post-GST):
p = 10/10,000 = 0.001
q = 1 - 0.001 = 0.999
f* = (227.50 x 0.001 - 0.999) / 227.50
f* = (0.2275 - 0.999) / 227.50
f* = -0.7715 / 227.50
f* = -0.00339
Kelly gives f* = -0.34% for the Starter Prize bet in isolation. The negative sign means: bet nothing, or if you can take the opposite side (be the lottery operator), Kelly says take 0.34% of bankroll on that position.
Aggregating across all prize tiers (as we did in the full EV calculation in 4D Lottery ROI Reality), the overall EV is -$0.398 per dollar staked. Kelly applied to the aggregate position gives:
b = 280 (simplified net odds)
p = 0.0022
q = 0.9978
f* = (280 x 0.0022 - 0.9978) / 280
f* = (0.616 - 0.9978) / 280
f* = -0.3818 / 280
f* = -0.00136 = -0.14%
Kelly says: the rational bet on a 4D ordinary draw, given these probabilities and payouts, is -0.14% of bankroll. The negative sign is unambiguous. The mathematically optimal amount of your bankroll to allocate to 4D lottery is zero.
What Kelly Tells Us — and What It Doesn't
Kelly's output of "bet nothing" in a negative-EV game is mathematically correct within its framework. But Kelly makes specific assumptions that are worth examining in the lottery context:
Kelly Assumes You Can Repeat the Bet Indefinitely
The Kelly framework's optimality proof relies on maximizing the long-run geometric growth rate of wealth across infinite repetitions. For 4D, if you participate for 30 years of weekly draws, you will approach the long-run expected outcome. If you participate for 20 draws, variance dominates and Kelly's recommendations for bankroll growth are less precise.
Kelly Assumes Wealth Maximization as the Objective
Kelly maximizes expected log-wealth — the mathematical representation of compound growth. This is the correct objective for an investment portfolio. It is not necessarily the correct objective for lottery participation, which many participants engage with for entertainment utility, not wealth maximization.
A participant who derives genuine entertainment value from lottery participation — the anticipation of draws, the social dimension, the occasional wins — is purchasing entertainment utility with their stake. Kelly does not account for utility from the participation itself, only for the financial outcome distribution.
Kelly Assumes Perfect Probability Estimates
In 4D, the probabilities are known precisely (regulated draw mechanics, published prize tables). Kelly's inputs are accurate. The output — don't bet — is therefore as reliable as Kelly ever gets.
The Fractional Kelly Adaptation for Lottery Participants
Professional bettors and traders commonly use "fractional Kelly" — betting a fixed fraction of the Kelly-optimal amount — to reduce variance at the cost of some expected growth rate. Typically, half-Kelly (betting 50% of f*) is used.
For negative-EV games, this framework inverts usefully: instead of asking "what positive fraction of Kelly should I bet?", we ask "what is the maximum fraction of bankroll I should risk given Kelly's verdict?"
The practical adaptation: treat the Kelly output as a ceiling on participation intensity rather than a bet size recommendation. Kelly says the rational stake is 0% of bankroll. The practical ceiling for a disciplined participant who accepts the entertainment utility argument is:
Not: Investment % of financial portfolio
Not: Income-replacement strategy
Practical ceiling (consistent with entertainment utility framework): 1-3% of monthly income
Practical Kelly-consistent floor: 0%
Any allocation within 0-3% of monthly income is defensible under the entertainment utility argument. Any allocation above 3% of monthly income begins to conflict with the math — you are either treating it as investment (which Kelly says don't) or allowing entertainment spending to crowd out financial obligations.
Per-Draw Kelly Sizing Within a Fixed Monthly Allocation
Once you have established a monthly bankroll allocation (say, $150/month), Kelly can be usefully applied at the per-draw level to determine unit sizing. Here, the relevant question changes from "should I bet?" to "given that I am participating, how much per draw preserves bankroll most effectively?"
Within a fixed allocation framework, the goal is to maximize the number of participation opportunities while respecting the monthly ceiling. This is equivalent to minimizing ruin probability for the fixed allocation.
The mathematically appropriate bet size for this objective — derived from ruin probability theory rather than Kelly directly — is:
For $150/month, 12 draws/month (3x weekly Singapore 4D), safety factor 2:
Optimal unit = $150 / (12 x 2) = $6.25 per draw
At $6.25/draw x 12 draws = $75 expected spend (with 2x safety factor for winning draw reinvestment)
The safety factor accounts for the fact that wins will be reinvested (or not, depending on your framework — see Bankroll Management Rule 2). A safety factor of 2 is conservative and appropriate for beginning participants. More experienced participants who maintain accurate records may reduce to a safety factor of 1.5 once their actual variance profile is understood.
System Bet Sizing Under Kelly Framework
System bets (covering multiple numbers) change the probability and payout structure significantly. A System 7 bet covering 35 numbers has probability 35/10,000 = 0.35% of hitting at least one prize tier. The payout per hit is correspondingly lower (prize divided across the cost of coverage).
Applying Kelly to a System 7 bet at $1 base coverage ($35 total):
b_system = (10 x $227.50) / 35 = $65 net per $35 invested (if exactly 1 of 35 numbers wins Starter)
p_system = 35 x (10/10,000) = 0.035 (probability of any Starter hit across 35 numbers)
q_system = 0.965
f* = (65 x 0.035 - 0.965) / 65 = (2.275 - 0.965) / 65 = 1.31 / 65 = +0.0202
This simplified calculation produces a positive Kelly fraction — but the simplification is significant. The calculation above considers only Starter Prize hits and treats the system as a binary bet. When all prize tiers are properly aggregated and the full cost structure is included, the overall EV of a System 7 bet remains negative (the same 40% house edge applies). The Kelly output returns to negative.
The takeaway: system bet complexity does not improve EV or change the fundamental Kelly verdict. It changes the distribution of outcomes — more frequent smaller wins, less frequent large wins — without affecting the long-run expected return.
Practical Bet Sizing: A Decision Table
| Monthly Bankroll | Draws per Month | Kelly-consistent Unit | Bet Type |
|---|---|---|---|
| $50 | 4 (weekly) | $4-6 | Ordinary Big, 1-2 numbers |
| $100 | 8 (2x weekly) | $5-8 | Ordinary Big, 2-3 numbers |
| $150 | 12 (3x weekly) | $6-10 | Ordinary Big, 2-4 numbers, or iBet |
| $300 | 12 (3x weekly) | $12-20 | Ordinary Big/Small, System 4-5 |
These figures are calibrated to leave a 50% buffer within the monthly allocation for variance — acknowledging that some months will see heavier-than-expected losses and the buffer prevents breaching the allocation ceiling.
What Kelly Cannot Do for You
Kelly, even as adapted here, cannot make 4D lottery profitable. It cannot identify winning numbers. It cannot improve the probability structure of any draw. What it does provide is a rigorous framework for:
- Confirming that the allocation ceiling should be entertainment budget, not investment capital
- Sizing per-draw bets to maximize participation duration within a fixed allocation
- Evaluating whether any structural changes (cashback programs, operator switching — see Cashback Optimization) meaningfully shift the EV calculus
- Providing a systematic, emotion-free framework for bet decisions
Combined with the payout calculation discipline from How to Calculate Real Payout and the behavioral rules from Bankroll Management, the Kelly framework completes a mathematically coherent approach to 4D participation — one that accepts the odds honestly and manages the experience rationally.