Sports Betting Mistakes That Cost You: Parlays, the Martingale Trap, and the Gambler's Fallacy

By · author of the MIT-licensed teachersbettextbook library ·

Four ways to lose money that have nothing to do with picking wrong. Average Joe is about to do all of them in one afternoon.

Meet your classmates

Same five. Today Joe is the lesson.

1. Coins have no memory

Chalk has lost six in a row. Joe is certain she is "due".

She is not. The seventh race does not know about the first six. A 50/50 shot that has lost six straight is still exactly 50/50, and believing otherwise is the oldest mistake there is. It has a name: the gambler's fallacy.

What is true is that a long run of losses stops looking unlikely once you count how many runs there were. Six in a row happens about once every 64 tries, and there are a lot of tries in a season.

2. Doubling up after a loss

Lose $1, bet $2. Lose that, bet $4. Sooner or later you win, and you are up a dollar. It is called the Martingale, and here is the uncomfortable part: it works. Most afternoons it works all afternoon.

Average Joe Bookie Brian +EV Evan Joe, doubling after every loss Evan, betting the same dollar $32 $0 $0.00 $33.00 Down $12 early, back up $19. Then one run too many. $0.00 $33.00 Down $10 early, back up $16. Then one run too many. $0.00 $33.00 Down $11 early, back up $7. Then one run too many. $0.00 $43.00 Down $15 early, back up $25. Then one run too many. $0.00 $35.00 Down $15 early, back up $15. Then one run too many. Lose $1, bet $2. Lose that, bet $4. It keeps working. Until five in a row, and there is nothing left to double. The strategy did not fail at the end. It was failing all day.
A real simulation, and one afternoon out of five picked at random. Both start on the dashed $32 line and bet the same coin-flip races at even money. Joe doubles after every loss and resets after every win, which is the Martingale. Watch the arc: an early cold streak digs him a hole, and the doubling digs him out and puts him properly ahead, exactly as promised. That recovery is the trap, because the same doubling meets one streak it cannot pay for. 1+2+4+8+16 is 31 of Joe's first 32 dollars, so a five-loss run takes everything he has whenever it lands, and one turns up about once every 32 races. Evan's flat dollar is quietly up here on nothing but the day's luck; a fair coin owes him exactly $0.

Watch Joe's line. Small steps up, over and over, for most of the day. Then five losses in a row and it is all gone in about ninety seconds.

1 + 2 + 4 + 8 + 16 is 31, and Joe started with 32. That is the whole trap: the ladder is always longer than the money, and five in a row turns up about once every 32 races.

3. Tying bets together

Three bets at −110 on one ticket pays about 6.96 to 1. That is a real number and it is why parlays feel good.

What the ticket does not say is that Brian's cut multiplies too. One −110 bet gives up about 4.5%. Three of them tied together gives up about 12.7%. You bought the same three opinions and paid the cut three times.

None of which makes a parlay illegal or stupid. It makes it an expensive way to buy a big swing, and the price of the swing is on the ticket if you work it out.

4. Betting more because you are down

This is the one that actually empties an account. Kelly says bet a fraction of what you have, so a shrinking bankroll means shrinking bets. Chasing does the exact opposite: it makes the bets bigger precisely as the money to cover them gets smaller.

It is not a strategy. It is Kelly, run backwards, on purpose.

None of these four is about picking the wrong hamster. Every one of them is about what you did after the race.

And one that is not a mistake, just uncomfortable

Twenty bets tell you almost nothing. A genuinely good bettor loses plenty of twenties, and a genuinely bad one wins plenty. If you want to know whether you are any good long before the results do, that is closing line value, and it is lesson 7.

Want more math?the Nerd Corner

Nerd Corner

Advanced material. Nothing above depends on it.

Why even the winning recovery bet loses

Worked example: why even the “recovery” win loses

The stake doubles after each loss:$25→$50→$100→$200→$400\text{The stake doubles after each loss:}\quad \$25 \to \$50 \to \$100 \to \$200 \to \$400
After 4 straight losses, already risked =$25+$50+$100+$200=$375\text{After 4 straight losses, already risked}\ = \$25+\$50+\$100+\$200 = \$375
The 5th bet, $400, finally WINS at −110:profit=$400×100110=$363.64\text{The 5th bet, } \$400\text{, finally WINS at } -110:\quad \text{profit} = \$400 \times \tfrac{100}{110} = \$363.64
Net=$363.64−$375=−$11.36(a win, and still down)\text{Net} = \$363.64 - \$375 = -\$11.36 \quad(\text{a win, and still down})
You won and you’re still $11.36 in the hole. A recovery win after nn losses nets base×(1−2n11)\text{base}\times\left(1 - \tfrac{2^{n}}{11}\right), negative the instant 2n2^{n} passes 11: exactly four losses (24=16>112^4 = 16 > 11). That, not the bankroll, is why the Martingale can’t dig back.

The four mistakes, priced

The parlay tax, drawn

The parlay product and the Martingale series

Each system collapses to one line of algebra. A parlay is a product, twice over: the combined decimal odds multiply, and, for independent legs, so do the win probabilities.

The parlay, as two products

Combined odds:d=∏idi=d1×d2×⋯×dn\text{Combined odds:}\quad d = \prod_{i} d_i = d_1 \times d_2 \times \cdots \times d_n
Fair prob (independent legs):p=∏ipi=p1×p2×⋯×pn\text{Fair prob (independent legs):}\quad p = \prod_{i} p_i = p_1 \times p_2 \times \cdots \times p_n
The payout multiplies posted prices; the fair probability multiplies true chances. The gap between those two products is the whole parlay tax.

Each posted price carries the book’s margin, so multiplying prices multiplies margins. One −110 leg implies 52.4% on a coin flip, a 4.8% markup over fair. Parlay two and the markup gets squared: the same expected value math, run in reverse.

Worked example: two legs at −110

One leg:d=100110+1=1.9091,implied 110210=52.4%\text{One leg:}\quad d = \tfrac{100}{110}+1 = 1.9091,\quad \text{implied } \tfrac{110}{210}=52.4\%
Parlay price:d=1.9091×1.9091=3.645  ⇒  implied 13.645≈27.4%\text{Parlay price:}\quad d = 1.9091 \times 1.9091 = 3.645 \;\Rightarrow\; \text{implied } \tfrac{1}{3.645} \approx 27.4\%
Fair (two coin flips):0.50×0.50=25%\text{Fair (two coin flips):}\quad 0.50 \times 0.50 = 25\%
EV per $1:0.25×3.645−1≈−$0.089\text{EV per } \$1:\quad 0.25 \times 3.645 - 1 \approx -\$0.089
Fair says 25%; the price implies about 27.4%. The per-leg markup squared (1.048×1.048≈1.0981.048 \times 1.048 \approx 1.098) is why the toll nearly doubles: −4.5 cents per dollar on one leg, −8.9 on two. In general, kk legs at −110 pay 0.9545k−10.9545^{k}-1 per dollar.

Run that out and the compounding is the whole story. Two lengths people actually play:

k=3:0.95453−1=−13.0%k=10:0.954510−1=−37.2%k=3:\quad 0.9545^{3} - 1 = -13.0\% \qquad\qquad k=10:\quad 0.9545^{10} - 1 = -37.2\%

Nothing about the legs got worse. The book took its 4.5% at each one, and 4.5% taken ten times running is 37%. The payouts are real. So is the rake, and it grows with every leg you add.

Any time a book quotes a parlay: convert each leg to decimal, multiply for the combined price, then multiply your own fair probabilities for the honest one. The difference is the fee: a number, not a feeling.

The Martingale’s one-liner is a geometric series. Stakes climb 1,2,4,8,…1, 2, 4, 8, \ldots base units, and the series sums to one less than the next power of two. The bankroll drains faster than the streak feels.

The Martingale stake series, summed

Total risked after n losses=1+2+4+⋯+2 n−1=2n−1 base units\text{Total risked after } n \text{ losses} = 1 + 2 + 4 + \cdots + 2^{\,n-1} = 2^{n} - 1 \text{ base units}
n=6:(26−1)×$25=$1,575(stakes $25→$800, next bet $1,600)n = 6:\quad (2^{6} - 1) \times \$25 = \$1{,}575 \quad (\text{stakes } \$25 \to \$800,\ \text{next bet } \$1{,}600)
Net if next bet wins at −110=base×(1−2n11),  negative once n≥4\text{Net if next bet wins at } -110 = \text{base} \times \left(1 - \tfrac{2^{n}}{11}\right),\ \text{ negative once } n \ge 4
At −110 the recovery profit is 2n×10112^{n} \times \tfrac{10}{11} base units against the 2n−12^{n} - 1 already lost; the 111\tfrac{1}{11} fee overtakes the +1+1 once 2n2^{n} passes 11, exactly four losses. Past that, the “recovery” win is itself a net loss: the trap the simulation above falls into.

Any staking plan that escalates after losses has a series like this underneath it. Sum it and check what it demands after an ordinary streak. That sum, not luck, is the system’s verdict.

Why twenty bets tell you almost nothing

The body says a twenty-bet record proves very little. This is the number behind that.

Flip a fair coin twenty times. How often does it land heads at least twelve, a 60% hot streak on a thing with no skill in it whatsoever?

P(X≥12)=∑k=1220(20k)(12)20=263,9501,048,576=25.2%P(X \ge 12) = \sum_{k=12}^{20} \binom{20}{k} \left(\tfrac{1}{2}\right)^{20} = \frac{263{,}950}{1{,}048{,}576} = 25.2\%

About one run in four. A coin does it. So whatever you concluded from your twenty bets, a coin would have handed you the same evidence a quarter of the time, and that is before anybody charges you vig on the twenty.

Expecting a short run to look like the long run has a name and a date: belief in the law of small numbers, 1971.[3] It is not a beginner mistake. The original paper was about working research psychologists doing it.

Where the psychology comes from

The arithmetic is a derivation you can re-run. The fallacy is one of the best-documented findings in the psychology of chance. The reading:

  1. Croson, R. & Sundali, J. (2005). “The Gambler’s Fallacy and the Hot Hand: Empirical Data from Casinos.” Journal of Risk and Uncertainty 30(3), 195–209. The fallacy on casino videotape, real money at stake.
  2. Clotfelter, C. T. & Cook, P. J. (1993). “The ‘Gambler’s Fallacy’ in Lottery Play.” Management Science 39(12), 1521–1525. Play on a lottery number falls sharply after it wins and takes months to recover.
  3. Tversky, A. & Kahneman, D. (1971). “Belief in the Law of Small Numbers.” Psychological Bulletin 76(2), 105–110. The small-sample illusion; the paper calls the gambler’s fallacy “a misconception of the fairness of the laws of chance.”

Every formula here lives on the Formula Sheet for quick reference.

The trap, run for a thousand bets

The scene up top is one 40-race afternoon. This is the long version: 1,000 coin flips at −110, the same flips for both lines. Martingale doubles after every loss and books its small wins right up until one streak is longer than the bankroll. The flat bettor just pays the vig slowly. Neither line is a strategy; that is the point. Replay it and the cliff moves, but it always comes.

Check your understanding

Frequently asked questions

Does the Martingale betting system work?

No. Doubling after every loss works until an ordinary streak demands more than your bankroll, and at −110 a recovery win after four or more losses is itself a net loss. The system busts about 96% of the time.

Why are parlay bets bad?

Each leg keeps the book's fee and the fees multiply: a three-leg −110 parlay carries about −13% expected value, a ten-leg one about −37%.

What is the gambler's fallacy in betting?

The belief that independent events self-correct, that a side is “due” after a streak. Coins have no memory.

Is chasing losses ever worth it?

No. Chasing raises your stake as your bankroll falls, the opposite of Kelly sizing, which stakes a fraction of what you currently have. It maximizes the odds of ruin.