Season 1 · Returns · Episode 1.1
Three answers
Twelve monthly returns. Add them: 12%. Multiply them: 11.54%. Add their logs: 10.92%. Three answers from the same twelve months, and only two of them are real returns. What each one is, the maths behind it (from the number e to the AM–GM inequality), which one you actually earned, why adding returns is always wrong, the two data checks to run before you trust any return, and the five mistakes that turn up in real reports. Every example starts at £100.
Plays here from YouTube, with English captions. Nothing is loaded from YouTube until you press play.
Chapters
Try it: twelve months, three answers
Change any month. The three answers update, and you can see which one is the money you actually made.
Adding is 0.46 points too high here. The bigger the swings, the more adding overstates: that is volatility drag, the subject of 1.2.
See it in APRI
Performance: returns linked by multiplying, gross and net of fees. The sandbox, to try it yourself on a sample client, is coming soon.
Sources
- Carl R. Bacon, Practical Portfolio Performance Measurement and Attribution, 2nd ed. (Wiley, 2008): return calculation, compounding, arithmetic and geometric means.
- CFA Institute, Global Investment Performance Standards (GIPS®), 2020 edition.
- NVIDIA, ten-for-one stock split, effective 10 June 2024.
- DAMA International, DAMA-DMBOK: Data Management Body of Knowledge, 2nd ed.: data quality dimensions (consistency, completeness).
Written, checked and animated by APRI Analytics. The narration is AI-generated. Education, not financial advice.
The script
Three answers, one portfolio
Here are twelve monthly returns from one portfolio. Add them up, and you get 12%. Multiply them, the way a bank balance would, and you get 11.54%. Take their logarithms and add those, and you get 10.92%.
Three different numbers, from the same twelve months, and only two of them are real returns. So which one is the return? By the end of this video, you'll know exactly what each of them is, and why only one of them is the money you made. All you need is percentages, and multiplication.
Oh, and keep that 10.92% in the back of your mind. It looks like the odd one out. It's going to turn out to be the most useful.
One period: price and total return
Let's start with one period. If you buy something for £100, and a year later it's worth £102, you get a price return: the change, divided by where you started. 2%. If it also paid you £3 of income along the way, a dividend, you add that in, and you get the total return. 5%.
If you add one to that, you get a growth factor: 1.05. It's simply what every pound turned into: £100 became £105. So: one period, one growth factor. But most of the time, you have more than one period.
Compounding: why (1+a)(1+b) isn't 1+a+b
First, the maths. Take two growth factors, 1 + a, and 1 + b, and multiply them. Picture a rectangle, 1 + a wide, and 1 + b tall. Its area comes in four pieces: 1, plus a, plus b, plus a × b. Add the returns, and you keep the first three pieces, but you throw away the corner, a × b. It's small, but it's always there. That's why returns multiply.
Now in money. If you have more than one period, you chain the growth factors. +10%, then −10%: £100 × 1.1 is £110. × 0.9 is £99. A 1% loss, not zero.
If you do that for every period, you multiply all the factors together. That's what the capital Pi means: a product. If you subtract one, you get the total return. For our twelve months, 11.54%. That's the number your account actually shows.
If you add the returns instead, you get 12%: a number nobody ever had in their account. So 12% is not what you earned. Adding returns is simply wrong. But that still leaves 11.54, and that mysterious 10.92.
How often? The number e
First, a famous piece of maths. Take 1 + 1/n, and raise it to the power n. When n is 1, that's 2. When n is 2, 2.25. 12 gives 2.61. A million gives 2.718. It keeps growing, but it never passes 2.71828. That limit is the number e. Expand the bracket, and it turns into 1, plus 1, plus a half, plus a sixth, plus a twenty-fourth, and so on, each term smaller than the last, and together they add up to e.
Now in money: how often is the interest paid? Say a bank offers 10% a year on £100. If they pay it once, at the end of the year, you get £110. If they split it into four quarterly payments of 2.5%, each payment earns interest on the ones before. You end the year with £110.38. Monthly: £110.47. Daily: £110.516.
The more often you compound, the more you get, but each step adds less than the last. If you keep splitting forever, (1 + r/m) to the power m stops growing. It settles at £100 × e^0.1: £110.517. A tenth of a penny more than daily. That limit is called continuous compounding.
And that's all the number e really is: what £100 becomes at 100% a year, compounded infinitely often. £271.83: e × £100. So growing continuously at a rate x turns £100 into £100 × e^x. Hold on to that, because it's the key to our mysterious third number.
Log returns: ln(ab) = ln a + ln b
First, the maths behind logs. The natural log answers one question: e to the what gives this number? And it has one superpower. Because e^x × e^y is e^(x+y), the log of a product is the sum of the logs. Logs turn multiplying into adding. That's how slide rules and log tables let people multiply big numbers by adding them.
Now in money. Here's the key. If £100 grows continuously at a rate x, it becomes £100 × e^x. So run that backwards. If £100 became £110, the continuous rate was the natural log of 1.10. That's called a log return. +10% is a log return of 0.095. −10% is −0.105. The loss is bigger, and now you can see it.
Now the superpower pays off. Logs turn multiplying into adding: the log of a product is the sum of the logs. So if you take our twelve months and add up their log returns, you get 10.92%. That's our mysterious third number. And it isn't a different answer. Put it back through e: e^0.10925 − 1, and out comes 11.54%. The same money, measured in different units. That's why professionals love log returns: they add up over time.
Across assets: the log curve's blind spot
First, the maths. Draw the log curve. It bends downwards. Pick two points on it, and join them with a straight line. The middle of that line is the average of the two logs. The curve above it is the log of the average. The line always sits below the curve. So the average of the logs is always less than the log of the average, unless the two numbers are equal.
Now in money, here's where that bites. Log returns have one blind spot. Take £100, and put half in something that makes 50%, and half in something that loses 50%. £50 becomes £75. The other £50 becomes £25. Together: £100. You made nothing. 0%.
Average the simple returns, +50 and −50, and you get 0%: right. Average their log returns instead, and you get −14.4%: wrong. A portfolio is a sum of pounds, not a product of factors, so across assets, it's simple returns that add up, weighted by how much you hold.
So here's the rule. Across time, growth factors multiply, so log returns add. Across assets, pounds add, so simple returns average. Mix them up, and a portfolio that went nowhere reports a 14% loss.
Arithmetic mean vs geometric mean
First, a classic inequality. For any positive numbers, the arithmetic mean is at least the geometric mean. Take 2 and 8. Their arithmetic mean is 5. Their geometric mean, the square root of 2 × 8, is 4. Picture it: a 2 by 8 rectangle, and a 4 by 4 square with the same area. The square's sides are the geometric mean. The two means are only equal when the numbers are equal, and the more the numbers spread out, the wider the gap. We'll prove it in the next video.
Now in money: which one should you quote? Take our twelve months again. The arithmetic mean, add them up and divide by 12, is 1% a month. The geometric mean answers a different question: what single monthly return, repeated twelve times, gives you the same money? Multiply the growth factors, take the twelfth root, subtract one: 0.915%, the compound return.
And here's the neat link. The geometric mean is just the arithmetic mean of the log returns, converted back through e. There's even a rule of thumb for the gap: the compound return is roughly the arithmetic mean, minus half the variance. Here that gives 0.914. Almost exact.
Now swing harder. +60%, then −40%. The arithmetic mean is 10% a year. But £100 becomes £160, then £96. You lost money. +50, then −50: an arithmetic mean of zero, and it costs you a quarter of everything. That gap is called volatility drag. The more returns swing, the more the arithmetic mean flatters you.
So the arithmetic mean describes a typical period. But it is not what an investor earned. For that, you need the compound return, from the geometric mean.
Who uses which answer
So, back to the start: three numbers, but only two real returns. Here's who uses them.
The compounded answer, 11.54%, is what you report. Factsheets, client statements, performance that follows the GIPS standards, and every 'per year' return over several years. That's what the money actually did.
The log answer, 10.92%, is the analyst's working unit. Risk models, volatility estimates, statistical tests and option pricing all use log returns, because they add up over time and behave well in the maths.
And the added answer, 12%? Nobody ever earned it, and no report should ever show it. Adding returns is simply a mistake. Two answers, two jobs. Compound your returns, never add them.
Data check 1: the crash that never happened
Before we trust any of these numbers, two checks on the data. Check one. Here's a share at £100. Overnight, its price is £50. A 50% crash? Look at what you own. Yesterday you had one share worth £100. Today you have two shares worth £50 each: still £100. The company did a two-for-one split. It cut every share in two, like cutting a pizza into more slices. Nothing was lost.
Real companies do this. When Nvidia split ten-for-one in June 2024, its share price went from about $1,200 to about $120 overnight. If your price data isn't adjusted, that's a 90% loss that never happened. A reverse split does the opposite: ten shares become one, the price jumps tenfold, and unadjusted data shows a 900% gain.
The fix is an adjustment factor. You divide every price before the split by the split ratio, so yesterday's £100 becomes £50, and the return is zero, which is the truth. It works because a return is a ratio, and scaling both prices by the same number leaves a ratio unchanged.
Dividends do the same thing on a smaller scale. On the day a £100 share stops carrying its £3 dividend, the ex-dividend date, its price drops to £97. The price return says −3%. But you have £97 of share and £3 of cash: still £100. And when a company spins off a division worth £20, the parent's price drops from £100 to £80. −20% unadjusted. Zero, once you count the new shares you were given.
So whenever a price moves more than the market could explain, check the corporate actions first. Data people call this consistency: every price in a series has to be on the same basis.
Data check 2: a gap isn't a zero
Second check. Our twelve months turn £100 into £111.54. Now suppose month twelve is missing: the file came in late, or a price feed failed. A good performance system stops right there and flags it. The danger is everywhere else. A price feed that fails often carries the last price forward, so December shows no change at all. Or someone compounds the column in a spreadsheet or a script, and the blank cell is quietly skipped.
Either way, the missing month is treated as zero, and the year comes out at 10.44%. A full point lower, and nothing on the report says anything is wrong. Treat a gap as zero, and you've invented a month where nothing happened. In maths terms, a zero return multiplies by one, the number that changes nothing. A missing month isn't that. A gap isn't a zero. It's a month you don't know.
So the rule is simple: count the months before you compound. Twelve months in a year, every one present, every one dated. If one is missing, the return isn't smaller. It's unknown, and the report should say so. Data people call this completeness: everything that should be there, is there. And it's one of the easiest checks to automate.
What goes wrong
Let's finish with the mistakes that turn up in real reports. One: adding monthly returns. 12%, instead of 11.54. Two: adding, or averaging, log returns across assets. Three: quoting the arithmetic mean as what investors earned. And then the two data checks. Four: unadjusted prices, so a two-for-one split looks like a 50% crash. And five: a missing month, quietly treated as zero.
Every one of these comes from forgetting the same idea: returns are growth factors, and growth factors multiply.
Back to the three answers
So, back to where we started. Twelve months, three answers. 12%: you get it by adding, and it's simply wrong. Nobody ever earned it. 11.54%: you get it by multiplying, and it's what you actually made. 10.92%: the same money, in log units. The version you're allowed to add across time.
Everything today came from one small idea. A return isn't a number you add. It's a factor you multiply: 1 + r. We saw one cost of that idea. +50%, then −50%: an arithmetic mean of zero, and it costs you a quarter of your money. That gap is volatility drag, and it isn't bad luck.
Next time, we'll prove it always happens, see what it does to leveraged funds, and learn the honest way to turn any return into a yearly one. When averages lie: that's next.