Nassim Taleb is remarkably brilliant. It's his work that's cited in the article. In my opinion, there are 2 choices, a misquote, if the article is wrong, or a misunderstanding on the part of the reader. There are a few things going on. Thanks to member Justin, I fixed the Wikipedia article link. I recall his assertion from the book "The Black Swan" (p275). And here it is -

and the referenced chart -

Now, with thanks to member Money Ann, who actually noted that the product of the 10 best days, was, in fact 64%.

Putting on my math hat, those ten days, cumulatively, multiplied one's wealth by 1.64. Game over. Had you 'not' been in the market the full ten days, it doesn't matter how far back you go, nor how far forward. Pull those numbers out and you have to divide your wealth by 1.64. (The only argument one might have is that, for example, deposits are made along the way, I, for instance, only started investing in 1984, so earlier numbers don't matter. That's a distraction, not the point of the long term observation).

To simplify my examples, say there was one day that the S&P went up 10% (for easy math). And we have the 4300% return long term that Money Ann cites. Remove that one day and you'd have only 3909% return. Not 4290%.

So, in fact, no surprise, the citation is accurate, although in the book, Taleb is more vague.

If my answer here needs any clarification, I am happy to do so. Please comment and I'll return, edit, and clean up comments.

Edit - in response to Dennis’ comment. Say there were a crash, and right after, the market recovered 50% in just one day. In a history of daily returns, we’d now have 1.5 as a factor. Now, over a long period of time, decades, we see the market up 1900%, i.e. the multiplication result is 20, as we flip from percents to factors. Remove the 1.5, and the result is simply 10, or growth of 900%. That one day, in or out, made a huge difference. It’s for the reader to keep an open mind, and realize it doesn’t take too many days to multiply to get that 50%. In fact, it’s not even 10. The moral of Taleb’s story is simply that trading, getting in and out of the market is a greater risk than staying in for the long term. (And note to Dennis - mhoran already made the same 50% math example. This is just the same in my own words.)

Since the book was written in 1997, the relevant period would be roughly 1947-1996. Yahoo Finance data starts from 1950, so I will look at "last 47 years" instead. Presumably, the finding by Mandelbrot and Taleb is not so trivial that it would no longer apply to even a slightly different time period.

On Jan 3, 1950 the S&P closed $16.66. On December 31, 1996, it closed $740.74. This is an increase of 4300% in total. The best days were:

1987-10-21 9.10%
1987-10-20 5.33%
1970-05-27 5.02%
1987-10-29 4.93%
1982-08-17 4.76%
1962-05-29 4.65%
1974-10-09 4.60%
1957-10-23 4.49%
1974-10-07 4.19%
1974-07-12 4.08%

This sums to 50%. Perhaps what is meant is that if you had only traded on the 10 best days you would make the lion's share of profit. To verify that we can take the product and obtain a 64% increase. So as far as I can tell, the claim as presented appears to be false. Neither Mandelbrot nor Taleb are generally considered fools or charlatans, so I'm assuming something was lost in translation. It is interesting that the 64% I got was very close to the 63% cited.

63% of 4300% is 2709% (if taking the percent of a percent confuses you, just think of making $4300 for every $100 you put in, and taking 63% of that). To obtain this profit by only trading on the best days, you would need to trade on about 115 days spanning from 1950 to 1991. Interestingly, if you traded on only the best days, your maximum gain would be about 3*10^12% from trading on 10240 of 11826 trading days in the dataset.

Incidentally, if you traded on 20 worst days, you would lose two thirds of your money. Your outlook only improves if you traded on the worst 10040 days, in which case you also lose two thirds. You can at most miss 140 of the best days if you want to at least break even, assuming you trade on every other day.

Interesting as it is to debate all this arithmetic, the second part of your question is more useful to discuss:

Or is it even fair to ask if this is true?

It doesn't really matter exactly what the numbers are. The authors' point is that there is an exponential distribution in stock market returns, where you make or lose huge sums on a minority of days, while most days don't really affect you one way or the other. Of course, the "long" or "fat" tails in market return distributions are well known. In nature, almost everything is distributed such that extreme events are rare, and common events are small. We do not have many phenomena where most values are very far away from the mean or median, the market is no exception. The debate that Mandelbrot and Taleb are addressing here is with regards to exact probability of a given unlikely event being 10^-9 vs. 10^-10 vs. 0. It may seem like it's academic, but in some cases it can make a big difference. The difference is unlikely to be apparent for a non-technically sophisticated trader.

Let’s simplify it, our example investment is very boring except for one day when it goes crazy.

Two investors with same $1,000 initial investment, the market doubles every 7 years except for one day at the end of the first 7 years when it goes up by 50%. The first investor invests for the whole period, the second investor skips just one day.

Investor 1:

• Starts with $1,000


• 7 years later $2,000 (2x)


• Crazy day with a 50% increase ends with $3,000


• 28 years later $48,000 (16x)

Investor 2:

• Starts with $1,000


• 7 years later $2,000 (2x)


• Misses the crazy day but then reenters the market still at $2,000


• 28 years later $32,000 (16x)

Investor one sees the initial investment go up 48x or 4700% Investor B sees the value go up by 32X or 3100%. So that 1 day with a 50% would erase about 50% of the gains even though the unlucky investor saw 3100% gain.

I did find that a similar phrase in an article on the motley fool website

Time in the market, versus time out of the market

J.P. Morgan Asset Management's 2019 Retirement Guide shows the impact
that pulling out of the market has on a portfolio. Looking back over
the 20-year period from Jan. 1, 1999, to Dec. 31, 2018, if you missed
the top 10 best days in the stock market, your overall return was cut
in half. That's a significant difference for only 10 days over two
decades!

The JP Morgan study can be found on their website. Page 41 of the report has information.

I can't speak to the research methods used in that study but Taleb was likely trying to build on his "black swan" hypothesis by showing that the "black swan" trading days have the biggest impact on the market overall.

The math behind Mandelbrot's and Taleb's analyses always goes over my head, even though I'm a fan of Taleb's work from a philosophical standpoint.

Tony Robbins simplified this concept by enforcing the idea that you can't time the stock market. I'm not sure who did the research, but it shows that if you try to time the market and miss out on the top performing days, you ultimately underperform the market.

Image source: MarketWatch

Perhaps true, but not evidence that returns are non-normal. I agree they are not normally distributed but disagree with the reasoning. As others pointed out, if the return of the top ten days is 63%, the claim is true regardless of the the performance of the other days. This can be achieved with a sufficiently volatile normal distribution. If the standard deviation of the returns is low and it is achieved, together it may be evidence of heavy tails. In isolation, it is not.

The simple script below generates normal returns such that the top 10 days have approximately a 70% return. Adjusting the standard deviation parameter will show the relationship.

import numpy as np

n_simulations = 5000
n_days_per_year = 250
n_years = 50
n_days = n_days_per_year*n_years
mean = 0.1/n_days_per_year
std_dev = 0.25 / n_days_per_year**0.5

rnds = np.random.normal(size=(n_days,int(n_simulations/2)))
rnds = np.concatenate((rnds,-rnds),axis=1) # antithetic

returns = mean + std_dev*rnds

sorted_returns = np.array([np.sort(returns[:,i_simulation]) for i_simulation in range(n_simulations)]).T
top_ten_returns = np.product(1+sorted_returns[-10:,:],0) - 1
print(np.mean(top_ten_returns))

I have seen many claims of this nature. I suspect that most of them have done the arithmetic correctly. What it is being used for here is to claim that the normal distribution is a bad model of reality far from the mean, because events many standard deviations out are much more common than the normal distribution would claim. The normal distribution is very convenient because we have lots of theorems about what happens when things are normally distributed. As long as you stay close to the mean it doesn't matter much which bell shaped curve you use. When you go far out it matters a lot. In real life the tails are always greater than a normal distribution will say.

The fact that the arithmetic is right doesn't tell us how to react to the fact. It is often cited to tell you not to try to time the market, because if you miss the best 10 days you miss so much of the return. On the other hand, if you miss the 10 worst days your return goes way up. No justification is offered that trying to time the market makes you more likely to miss the great days than miss the terrible days.

Lots of great mathematical exegesis here but looking more widely there seem to be two primary motivations that people have for making these comments about "best days".

One is the Nick Taleb point that 'fat tails' mean that the distribution of returns is concentrated (for good or ill) in a few incidents. This seems valid.

You will also come across it as an argument against pulling out of the market in periods of volatility. This is not really a valid argument - it rather ignores the point that if you instead caught the "worst days" of the market you would be suffering huge losses. (which is not to say the conclusion is wrong - just the argument)

(FYI, if you are interested in long-term stock market returns the go-to-guy is Elroy Dimson - a professor at LBS & the Judge Instittue. He has analysed this across multiple countries and periods).

tl;dr- This claim is false. While it's true that an investor who bought in at the start of each of the top-10 days, then sold at the end of them, could enjoy gains of 64% over a similar investor who didn't participate on those 10 days, it doesn't follow that these top-10 days "represent 63 per cent of the returns of the past 50 years".

The claim is incorrect due to a broken frame.

Imagine that someone tells you that the best strategy for your financial future is to use a certain strategy for playing poker at a Las Vegas casino. They might be correct about their strategy being optimal, in a frame that assumes that you're going to be betting your financial future at a casino. But, is it then true that the optimal strategy for your financial future involves playing poker in a Las Vegas casino?

It's not that the claim about how to play poker is necessarily wrong, but the frame around it is. Likewise, the title of this question asks:

Is it true that, “just ten trading days represent 63 per cent of the returns of the past 50 years”?

That "63 per cent" figure may come from correctly performed mathematical operations, but the frame suggests that someone who missed those 10 days could only hope to enjoy the remaining 37% of returns over the past 50 years.

The problem's that it's generally inappropriate to claim that members of some set constitute some portion of that set's total when that set isn't just adding over positive values. (More about the math at the end of this answer.)

For example, going by Wikipedia's numbers, it looks like an investor who skipped the worst-5 days would've applied a ~79.4% gain to their wealth. But does it make sense to attribute ~79.4% of wealth to skipping the worst-5 days while also attributing 63% of the wealth to being present for the best-10 days?

Analogy: Did one employee do 90% of a small operation's work?

Consider a small company with 25 employees. They do little contract jobs.

One employee, Bob, has a minor role as a specialist on 90% of the projects; the other 10% don't involve his area of expertise. Bob then writes on his resume:

My efforts represented 90% of the company's productivity during my tenure.

Which makes it sound like he did 90% of the work, and everyone else did 10%, right? Clearly he's a top-notch employee!

And he's not lying, in the sense of making numbers up. That figure's accurate, for the frame in which it was calculated.

The problem's that the frame's broken. This is, it doesn't make sense for Bob to describe his participation in 90% of the projects as having represented 90% of the company's work. It's not that he did the math wrong, it's that the logic behind the model itself is a tad silly.

Corrected claim: Investing on the top-10 best days would yield a ~64% return.

It seems that, if an investor invested at the start of a day and then pulled out at the end for each of the top-10 days, they'd have applied a factor of ~164%, for a gain of ~64%.

Without checking the math or figures, that looks plausible to me. But, assuming it's true, it becomes false if we misrepresent that claim through wording that makes it sound like this is a thing where it's meaningful to attribute 64% of gains to 10 specific days, such that anyone who missed those 10 specific days can only hope to enjoy 36% of gains, or something like that.

It's not that someone did the math wrong or messed up calculating the figure; rather, it's that the model frame in which the figure exists is itself broken. The correct solution is to exit that frame.

Discussion: Mathematical explanation.

When we say that something contributes x% to a total, we're typically working within a framework in which we're adding a bunch of positive values to arrive at a total.

We can make these investments additive. Specifically, we can take the natural-logarithm of each day's ratio to arrive at an additive parameter. Then, we can sum those additive parameters up, raise Euler's number to the power of their sum, and that's the total factor.

To grab the table from @MoneyAnn's answer, then augment it:

 Date +% factor ln(factor)
-----------------------------------------------------
1987-10-21 0.0910 1.0910 0.0870947068509337
1987-10-20 0.0533 1.0533 0.0519280928603591
1970-05-27 0.0502 1.0502 0.0489806222216219
1987-10-29 0.0493 1.0493 0.0481232751817282
1982-08-17 0.0476 1.0476 0.0465018336514199
1962-05-29 0.0465 1.0465 0.0454512629039174
1974-10-09 0.0460 1.0460 0.0449733656427312
1957-10-23 0.0449 1.0449 0.0439211870579281
1974-10-07 0.0419 1.0419 0.0410459694360010
1974-07-12 0.0408 1.0408 0.0399896482161584
 ------------------
 Total: 0.498009964

Then exp(0.498009964) is 1.645443519, which recovers that ~1.64-factor, or ~64%, other answers were referring to.

However, the broken frame presented in the claim has 2 major problems:

1. The factors aren't totaled over all days.
   If someone wanted to make a claim like this, then they should've ln(factor)'d all of the days, then divided the sum of those parameters for the top-10 days by the sum of those parameters for all of the days.




2. The factors contain negatives.
   Some days ended lower than they started. For example, apparently 1987-10-19 had a −20.47% hit, for a ln(factor) ~= ln(1-0.247) ~= -0.231553819. In general, for any factor of less than 1, ln(factor) should be negative.

We could try to fix up the numbers to account for Problem #1, but the frame doesn't lend itself to resolving Problem #2.