Bankroll and variance

Two numbers describe different things

An expected loss of a few pounds can sit beside a session loss of a hundred pounds without any contradiction. Expectation describes the average over repeated comparable play. Variance describes the spread of possible results around that average. Neither tells you the next card.

A bankroll is money set aside for play. For recreational gambling, treat it as a spending limit you can afford to lose, rather than capital certain to earn a return. A larger bankroll can let you absorb more swings; it does not turn a negative-expectation game into a positive one. There is no universally safe number of betting units.

The basic-strategy chart addresses playing decisions. Your budget addresses what a losing session would mean outside the game. Those questions deserve separate answers.

Expected cost uses the amount put at risk

Here is an illustration, not a measured house edge for our trainer. Assume a game has a 0.5% house edge relative to initial wagers, and you play 100 rounds at a £10 initial stake:

100 × £10 × 0.005 = £5 expected loss.

This is not a £5 session fee or a promise to lose exactly £5. You could finish ahead or lose much more. If the average initial stake becomes £25, the same hypothetical calculation gives £12.50. If you play twice as many rounds, the expected cost doubles.

Use the denominator supplied with a house-edge estimate. An edge quoted per original wager already accounts for the strategy’s doubles and splits; adding those wagers again to the same formula would count their effect twice. Other return measures may use total money wagered, so read their definitions. The Wizard’s blackjack variance reference reports expected values and variance for specified rules rather than one universal blackjack figure.

Put a scale on the swings

That reference gives per-round variance near 1.3 squared betting units for several basic-strategy games. The square root is about 1.14 units of standard deviation. For an explicitly simplified model, suppose rounds are independent, variance is exactly 1.3, bets stay at £10 and the hypothetical edge remains 0.5%.

Standard deviation after n rounds is £10 × √(1.3n):

These are scales of variation, not minimum or maximum losses. The table assumes unlimited ability to continue, fixed bets and independent rounds. Actual finite-shoe play, rule differences, changing stakes and stopping when funds run out complicate the distribution. Do not use it to assign an exact probability to your own session.

The central insight survives those qualifications: in this model expectation grows in proportion to the number of rounds, while standard deviation grows with its square root. More play increases absolute swings even while averages become more informative. “The long run” does not mean the next evening must repay last evening’s loss.

One round can require several stakes

A £10 minimum does not mean a £10 maximum exposure per round. A double adds another £10. Splitting begins another £10 hand; subsequent doubles may add still more. Under a four-hand split limit with doubling permitted on all four resulting hands, the theoretical total exposure is eight initial units, or £80 from a £10 starter. Some hands cannot reach that pattern, and split aces have special restrictions.

The mechanics are explained in doubling down and splitting; the blackjack rules reference describes the additional wagers. Before placing a stake, leave room within your spending limit for the legal decisions it may create. If the table minimum makes that uncomfortable, the minimum is too large for that session budget.

Why “double after every loss” fails

Start a loss-recovery progression at £10. Six consecutive losses cost:

£10 + £20 + £40 + £80 + £160 + £320 = £630.

The next proposed bet is £640. A single ordinary win there would recover £630 and earn £10, which explains the progression’s appeal. But the sequence needs £1,270 available to place every wager through that seventh bet, and a table limit can stop it earlier. Doubles, splits, naturals and pushes also make actual blackjack messier than this even-money sketch. Shackleford’s Martingale analysis explains the finite-bankroll failure.

Past losses are money already spent, not a debt the cards owe you. Moving a large stake onto the next hand creates a large new risk. A frequent small session win can coexist with occasional severe losses; win frequency alone does not measure a method’s expectation. The betting-systems analysis makes that distinction explicit.

Judge decisions before results

Set your spending and time limits before playing; stop when either is reached. Such limits control exposure and make a session manageable. They do not improve the probabilities or guarantee that the limit will never be lost.

In our trainer, practice without treating a streak as evidence of skill. Ask whether you identified the hand, checked the rules and chose the recommended available action. Learn the vocabulary of expectation, variance and house edge in the glossary, and compare rule quality in choosing a table. Good decisions are observable immediately. Their financial consequences remain uncertain.