Risk & Probability Models: Betting Without Fooling Yourself

Money decisions are bets. Not just the obvious ones at a poker table or a sportsbook — every choice about an uncertain future is a wager on what’s likely to happen. Buy a stock, take a job, insure your car, lend a friend cash: you’re staking something now on an outcome you can’t see yet. The good news is that humans have built a small toolkit for thinking about bets clearly instead of letting fear and excitement drive. That toolkit is probability — and this lesson hands you the six mental models that matter most. Master them and you stop reacting to the loudest outcome and start playing the long game.

Expected value — the long-run average, not the loud outcome

Picture a carnival game: pay $1, and one time in ten you win $5. Your gut hears “win $5!” and reaches for a coin. The carnival hears something quieter and far more profitable: on average, every $1 you spend hands them back $0.50. Same game, two completely different readings — and only one of them is doing the math.

Expected value (EV) is that math. You take every possible outcome, multiply each by its probability (how likely it is, written as a decimal from 0 to 1), and add them all up. The result is the average you’d earn per play if you could run the bet thousands of times. Formally, for a bet that wins or loses:

$\text{EV} = (P_{\text{win}} \times \text{win amount}) - (P_{\text{lose}} \times \text{loss amount})$

Worked example — the carnival game, decoded

You pay $1 to play. There’s a 10% chance (probability 0.10) you win $5, and a 90% chance (0.90) you win nothing and lose your $1:

$(0.10 \times 5) - (0.90 \times 1) = 0.50 - 0.90 = -0.40$

The expected value is −$0.40 per play. Every single game, on average, you’re handing the carnival 40 cents. Play once and you might walk off with $5 and a grin. Play a hundred times and the average grinds you down to a near-certain loss of about $40. The vivid $5 is real — it’s just rare, and the boring 40-cent leak is what actually decides your fate.

Drag the dials below. Watch how a juicy headline payout can still carry a negative EV the moment the odds turn against you — and how a modest, likely win quietly beats a flashy, unlikely one.

When to reach for it

Any time you face a repeated decision under known odds — insurance, a business with many small transactions, a portfolio of bets, a habit you’ll repeat thousands of times. EV is the lens for “what happens on average, over many tries.” (Its one blind spot: a bet you only get to make once, where a single catastrophic loss ends the game — that’s what the last section is about.)

Risk vs uncertainty — knowing the odds vs not even knowing the outcomes

Here’s a distinction that sounds like word-splitting until it saves your bank balance. A dice game and a brand-new startup are both “uncertain,” but in profoundly different ways — and treating one like the other is how smart people blow up.

Risk is when you can estimate the odds. You may not know the result, but you know the full menu of outcomes and roughly how likely each is. A roll of a fair die: six outcomes, each with probability 1/6. A coin flip, a roulette wheel, a well-studied medical procedure — the dice are unknown but the distribution is known. You can compute an EV.

Uncertainty (sometimes called Knightian uncertainty, after economist Frank Knight) is when you can’t even list the outcomes or their probabilities. What will a never-before-seen technology do to an industry over twenty years? What’s the probability of a genuinely novel crisis? Nobody knows, because there’s no track record to count. There is no reliable distribution to plug into an EV formula.

Why pretending uncertainty is risk is dangerous

The trap is false precision: dressing up a wild guess in the costume of a calculation. When someone hands you a tidy “there’s a 73.4% chance this works,” ask where that number came from. If it’s a die roll, trust it. If it’s a one-of-a-kind event with no history, that decimal is theater — a number invented to feel rigorous. The 2008 financial crisis was, in part, models treating deep uncertainty (will house prices ever fall everywhere at once?) as if it were tidy, well-measured risk. The math looked airtight; the inputs were fantasy.

Base rates & regression to the mean — start from how often, in general

A friend tells you, eyes wide, about a guy who quit his job, day-traded for a year, and made a fortune. Thrilling story. Useless data — unless you also ask: out of everyone who tried that, how many ended up like him? That second number is the base rate, and ignoring it is one of the most expensive mistakes in finance.

A base rate is how often something happens in general, across the whole population, before you zoom in on one vivid case. Most active day-traders lose money over time — that’s the base rate, and it should be your starting point. The lucky winner is a real story, but he’s one survivor pulled from a graveyard of people you never hear about. Always anchor on the base rate first, then adjust for what’s genuinely special about the specific case.

Worked example — the “90% accurate” test for a rare disease

Suppose a disease affects 1 in 1,000 people. A test is “90% accurate”: it catches 90% of sick people, and gives a false positive for 10% of healthy people. You test positive. How worried should you be? Most people guess “90% chance I’m sick.” Watch what the base rate does to that. Imagine 100,000 people:

Of everyone who tests positive, only about 90 are truly sick: $90 \div 10{,}080 \approx 0.009 = 0.9\%$ So a “positive” on a 90%-accurate test means roughly a 0.9% chance you’re actually sick — not 90%. Why so low? Because the disease is rare (the base rate is tiny), the flood of false positives from the huge healthy group drowns out the handful of true positives. Skip the base rate and you’d panic at a 1-in-110 risk as if it were near-certain.

Regression to the mean — extremes don’t last

Now the sibling model. Regression to the mean says that after an extreme result — wildly good or wildly bad — the next result tends to land closer to average. The star fund that crushed the market this year probably had a slug of luck baked in; next year it likely drifts back toward ordinary. This fools us constantly: a struggling player gets a stern lecture, then improves — and the coach credits the lecture, when the player was just regressing toward his normal level anyway. We credit and blame causes that did nothing; the real driver was math.

Asymmetry & convexity — small capped downside, large open-ended upside

Imagine two coupons. Coupon A: lose $10 for sure, but a 1% chance to win $5,000. Coupon B: win $10 for sure, but a 1% chance to lose $5,000. Same numbers, mirror-flipped. One is a gift; the other is a time bomb. The difference is asymmetry — and learning to feel it in your bones is one of the highest-leverage instincts in finance.

A payoff is convex when your downside is small and capped (you can only lose a little) while your upside is large and open-ended (you might win a lot). Coupon A. A payoff is concave when it’s the reverse: small, capped gains while the losses run open-ended. Coupon B — the “picking up pennies in front of a steamroller” shape. A linear payoff is the boring middle: gains and losses scale evenly, no curve either way. The mental-model rule is blunt: load up on convex bets; run screaming from concave ones.

Drag through the payoffs below to see the curve. The convex line barely dips on the downside but rockets up on the upside; the concave one inches up then falls off a cliff.

Everyday shapes: a lottery ticket is convex (lose a couple bucks, maybe win millions), and selling insurance is concave (collect small steady premiums, but one giant claim can ruin you). Convex = “limited loss, unlimited gain.” Concave = “limited gain, unlimited loss.”

The crucial catch — convex shape isn’t enough

Here’s where beginners get burned: a convex shape doesn’t automatically mean a good bet. The lottery is gloriously convex and still a terrible deal, because its EV is deeply negative (you remembered the −$1.67 ticket). The shape tells you about the spread of outcomes; the EV tells you whether the odds are priced in your favor. The dream bet is both: convex and with odds that aren’t stacked against you — a small, sure cost for a shot at a payoff the world has underpriced. Shape and odds are two different questions; always ask both.

Margin of safety — build in room to be wrong

Engineers don’t build a bridge to hold exactly the traffic they expect. They build it to hold several times that — trucks, a freak crowd, a storm, plus their own estimation errors — because being wrong about a bridge is not the kind of mistake you get to fix afterward. That buffer between what it can take and what it’ll probably face is the margin of safety, and it’s just as vital in money as in steel.

A margin of safety means deliberately building in a cushion so that being wrong doesn’t hurt you. You don’t size your decision to the rosy best case; you size it so that even a disappointing outcome leaves you fine. The bigger your uncertainty (remember that model two sections ago), the bigger the cushion you demand. It’s the financial version of “measure twice, cut once” — except you also leave the plank a little long on purpose.

Worked example — pay $70 for something worth $100

Say you reckon something is worth about $100 — a stock, a used car, a small business. You’re not certain; your estimate could be off. So instead of paying $100, you refuse to pay more than $70. That $30 gap is your margin of safety:

$100 − $70 = $30 cushion, a 30% margin of safety.

Now suppose you were too optimistic and it’s really only worth $85. Paid $100, you’d be underwater by $15. Paid $70, you’re still ahead by $15 — your error got absorbed by the cushion instead of by your wallet. The margin of safety doesn’t make you right; it makes being wrong survivable.

Ergodicity & never risk ruin — the average can’t save you if you’re out of the game

Time for the model that overrides all the others. Imagine 100 people each play one round of Russian roulette for $1 million. On average across the group, this looks amazing: roughly 83 walk away a million richer, only about 17 don’t walk away at all, so the “average payoff” per person is a fat positive number. Now imagine you play it 100 times in a row. Your average outcome isn’t a pile of money — it’s a near-certain end of the game, because all it takes is one fatal round and there’s no round 101 for you.

That gap has a name: ergodicity (and the lack of it). A situation is ergodic when the average across many people equals the average for one person over time. Lots of money situations are non-ergodic: what happens to you, sequentially, over time can look nothing like the rosy average across a crowd — specifically when one bad outcome ends the game. The crowd average quietly ignores that the dead players don’t get to keep betting. You don’t live as a crowd; you live as one person, in sequence, and you only get to keep playing if you don’t get knocked out.

The iron rule — avoid anything that can wipe you out

So the rule that trumps a good EV: never make a bet that can ruin you, no matter how good the average looks. “Bet the farm” is a warning, not a strategy. A 90%-chance-to-double, 10%-chance-to-lose-everything bet has a tempting EV — but run it enough times and the 10% will eventually land, and when it takes everything, you’re done. There’s no recovering from zero; you can’t compound from nothing, and you can’t make back a bet you’re no longer alive (financially) to place.

This is exactly why the earlier models matter so much together:

• Margin of safety keeps any single mistake from reaching the “wipe-out” zone.
• Asymmetry is your friend here — convex bets cap the downside, so a loss stings but never ends the game.
• Expected value is necessary but not sufficient: a great average means nothing if the path to it runs through a cliff you can fall off.

Putting it together

Six models, one job: bet on the future without lying to yourself. EV tells you the long-run average; risk-vs-uncertainty tells you whether your odds are real or invented; base rates and regression keep you anchored to how the world usually behaves; asymmetry tells you to chase capped-downside, open-ended-upside shapes; margin of safety leaves room to be wrong; and never-risk-ruin makes sure no single bet ends the game. Here’s the whole toolkit in one picture.

A mixed recap — it pulls from every section above, plus a callback to lesson one.

Key Takeaways