Additive vs Multiplicative

Lessons 01 and 02 handed you the two averages and made a coin misbehave. This lesson finds the fault line underneath all of it — the single structural question that decides, in advance, whether the ensemble average is your friend or a con artist. It comes down to one word: does your wealth add or does it multiply?

That fork is the whole game. On one side of it, the two averages agree and expected value tells the truth. On the other, they split, and the number your intuition trusts is computing somebody else’s future. And at the very end there’s a small mathematical miracle — a logarithm — that turns the dangerous side back into the safe side, which is exactly the move lesson 04 will optimise.

Additive dynamics: the swings that don’t care where you are

Picture a game where every round you either pocket a fixed $10 or pay a fixed $8 — and the amount never changes with your bankroll. Rich or broke, a win is exactly $10 and a loss is exactly $8. The step is blind to your current wealth. That blindness is the whole point.

Formally, an additive process updates your wealth $W$ by adding an independent random amount each round: $W_{t+1} = W_t + X_t$, where the size of $X_t$ doesn’t depend on $W_t$. Because the increments simply pile up, after $n$ rounds your wealth is your starting stake plus a sum of $n$ draws — and the long-run per-round change is just the ordinary arithmetic mean of those draws. Sums are the natural home of the arithmetic mean, and the arithmetic mean is exactly what expected value computes.

Here is the additive coin worked out. It’s a fair 50/50: heads pays +$10, tails costs −$8. The expected value per round is:

$\tfrac{1}{2}(+10) + \tfrac{1}{2}(-8) = +1$

A clean +$1 per round. Now the crucial claim: a single player, grinding this coin over time, also nets about +$1 per round. Watch one plausible run:

Six rounds, three heads and three tails, net change +$6 — that’s +$1 per round, precisely the expected value. The order didn’t matter (addition commutes: rearrange the flips and you land on $106 either way), and no single loss could scale up to swallow you, because a loss is forever just $8. The time average equals the ensemble average. This game is ergodic, and expected value is telling you the exact truth about your own future.

When to use it

Reach for additive thinking — and trust expected value outright — whenever the outcomes are fixed, bounded amounts that don’t scale with your total: a $5 side bet among friends, a rounding error, a small fixed fee, a coin for pocket change. If no single step can ever grow large enough to end the game, the crowd’s fate really is your fate.

Multiplicative dynamics: the swings that scale with the path

Now the version that runs the real world. Instead of adding dollars, each round multiplies your wealth by a factor: heads is ×1.5 (up 50%), tails is ×0.6 (down 40%) — the flagship coin from lessons 00–02. The step is no longer blind to your wealth; it is a percentage of wherever you currently stand, so its dollar size grows and shrinks with you.

Formally, a multiplicative process updates wealth by multiplying by an independent random factor: $W_{t+1} = W_t \cdot R_t$. After $n$ rounds your wealth is your stake times a product of $n$ factors, $W_n = W_0 \cdot R_1 R_2 \cdots R_n$ — not a sum. And products behave nothing like sums: they are order-insensitive in value but path-sensitive in consequence, a single tiny factor can gut the whole chain, and a factor of exactly zero is an absorbing barrier that no later win can undo.

Put the two coins side by side and the fork snaps into focus:

Same 50/50 fairness, same “positive edge” story on paper — and yet the additive coin gently enriches every patient player while the multiplicative coin, as lesson 02 showed, grinds almost everyone toward zero. The only thing that changed is add versus multiply. That is the fork, and everything downstream in this course hangs on which side of it you’re standing.

The master illusion: +50% and −40% do not net +10%

Here is the trap that catches nearly everyone, expert and novice alike — the master illusion of multiplicative games. Read the coin’s two moves — “up 50%, down 40%” — and your additive intuition sums them: $50 - 40 = +10$, so surely one win and one loss leave you up 10%. It feels airtight.

It’s wrong, and it’s not close. Percentages don’t add across rounds; the factors multiply:

$1.5 \times 0.6 = 0.9$

One win and one loss leave you at ×0.9 — down 10%, not up 10%. A twenty-percentage-point error, in the wrong direction, from a sum your gut swears is obvious. The additive mind sees $+50\% + (-40\%) = +10\%$; the multiplicative reality is $\times 1.5 \times 0.6 = \times 0.9$.

The asymmetry has a brutal, practical corollary. Because a −50% loss cuts you to half, undoing it takes not a +50% gain but a +100% gain — you have to double what’s left just to get back to where you started. Losses on the multiplicative side are structurally harder to reverse than same-sized gains are to hold, and it gets worse the deeper you fall:

The recovery column doesn’t just outrun the loss column — it accelerates. That runaway is volatility drag in disguise, and it’s why big swings quietly bleed a compounding bankroll even when the “average” percentage looks flat.

The log transform: how multiplication becomes addition

Everything above makes multiplicative games sound like a different, scarier universe with its own broken arithmetic. Here’s the beautiful part: one transformation turns that universe back into the safe, additive one. Take logarithms.

The whole trick rests on the defining property of the logarithm: it converts multiplication into addition.

$\ln(a \cdot b) = \ln a + \ln b$

So apply $\ln$ to a multiplicative process and the product of factors becomes a sum of log-factors: $\ln W_n = \ln W_0 + \ln R_1 + \ln R_2 + \cdots + \ln R_n$. In log-wealth, the process is purely additive — the very structure we already proved is ergodic. And because it’s now a sum, the honest long-run growth rate is just the ordinary arithmetic mean of the log-returns. The scary multiplicative game and the tame additive game are the same game viewed through two different lenses; the logarithm is the lens that makes the time average computable.

Let’s derive the flagship coin’s true growth rate from scratch — the −5% that lesson 02 asserted, now earned. The two factors are 1.5 and 0.6. Their log-returns:

• $\ln(1.5) \approx +0.405$
• $\ln(0.6) \approx -0.511$

Average them (arithmetic mean, because in log-space it’s additive):

$\tfrac{1}{2}(0.405) + \tfrac{1}{2}(-0.511) = -0.053$

That −0.053 is the average log-return per round. Convert back out of log-space to get the actual per-round wealth factor:

$e^{-0.053} \approx 0.949$

A per-round multiplier of ≈ 0.949 — about −5% per round, exactly the time-average decay lesson 02 measured, now derived from first principles. And notice what $e^{-0.053}$ is: it’s the geometric mean of 1.5 and 0.6, since $\sqrt{1.5 \times 0.6} = \sqrt{0.9} \approx 0.949$. That’s the deep reason the geometric mean (the average taken in log-space) is the honest per-round factor for multiplicative growth, while the arithmetic mean of the raw factors (the +5% ensemble number) answers a different question entirely.

Why a small side-bet is almost additive: the bridge to Kelly

If multiplication is the poison, here’s the antidote in embryo: stake only a tiny fixed fraction of your wealth, and a multiplicative game starts behaving like an additive one again. The two averages, torn apart by full-stake compounding, quietly drift back toward each other.

Two things happen when you shrink the bet. First, losses stop being ruinous: if you risk 2% of your wealth per round instead of 100%, no single flip — or even a long cold streak — can push you near the absorbing barrier at zero, so the game never actually ends and you’re always around to collect the average. Second, and more subtly, over a small range the logarithm is nearly a straight line: for tiny moves $x$, $\ln(1+x) \approx x$. In that near-linear zone, “the average of the logs” and “the log of the average” almost coincide — so the geometric mean and the arithmetic mean nearly re-converge, and the gap between the time average and the ensemble average shrinks toward zero.

Concretely: our full-stake coin costs about −5%/round (geometric mean 0.949). Cut the stake to a small slice of wealth and the per-round drag collapses toward — and past — zero, turning that same fair coin from a slow bankruptcy into a genuine engine of growth, without touching the odds. You didn’t fix the coin; you dragged a non-ergodic gamble back toward the ergodic side of the fork by refusing to bet the whole path on any one flip.

That is the exact hinge lesson 04 walks through. Kelly betting is precisely the recipe for how small — the fraction that maximises the time-average (log) growth rate you actually live, rather than the ensemble average that lures everyone else onto the rocks. This lesson found the disease (multiplication) and named its cure in principle (shrink the bet, restore near-additivity); Kelly turns that principle into a number.

Pitfall: wearing additive glasses in a multiplicative world

The single most expensive mistake in all of finance and gambling is importing additive intuition into a multiplicative game — and you’ve now met its flagship form. “Up 50%, down 40%, so I’m up 10%” feels like arithmetic because your brain was trained on additive childhood examples (allowances, marbles, fixed bets) where it was genuinely true. It stops being true the instant outcomes are percentages of a moving balance, because then each move multiplies and $1.5 \times 0.6 = 0.9$, not $1.1$.

The same error wears other costumes. “My fund averaged 10% a year” usually quotes the arithmetic mean of annual returns, which systematically overstates the growth an investor actually compounded — the geometric mean is always lower, and the gap widens with volatility. This is volatility drag: two funds can share an identical average return, yet the choppier one leaves you poorer, because its bigger swings multiply through a lower geometric mean. Whenever someone reasons about rates, returns, or percentages that stack round after round using plain addition, they are quietly computing the ensemble average and mistaking it for their own path.

When to use it: the one question to ask before you average

The entire lesson collapses to a single reflex you can run in a heartbeat, before trusting any average anyone hands you:

Does this outcome ADD to my total, or MULTIPLY it?

• If it adds a fixed, bounded amount that doesn’t scale with your wealth — the process is additive, ergodic, and expected value (the arithmetic mean) is telling you the exact truth. Trust it.
• If it multiplies — a percentage of your current stake, a return, a growth rate, anything that compounds — the process is multiplicative and non-ergodic. The arithmetic mean is the crowd’s number, not yours; take the geometric mean (average the logs) for the rate you’ll actually live, and remember that a step near zero can end the game forever.

That fork — add or multiply — decides which average is telling the truth, and asking it costs you nothing. Getting it wrong has bankrupted people who could compute expected value in their sleep.

Recap

Next up: lesson 04, The Kelly Fix — we take the “shrink the bet” cure this lesson named in principle and turn it into a number. Kelly betting maximises the time-average (log) growth rate you actually live, finding the exact fraction that keeps you on the growing, ergodic side of the fork instead of the crowd-flattering, path-destroying one.