I have written about the topic of volatility clustering a number of times, and thought I would update the post here. Many people trot out the highly misleading statistic about missing the best 10 days as a reason excuse for buy and hold investing. It drives me crazy to the point I have considered writing a paper on the topic. Great background reading here from Gire and Estrada. Stay tuned.
However, this line of reasoning is way too simplistic. The basic math shows that the vast majority of up AND down days occur when the market has already been declining. The simple reason? The market is more volatile. The last eight months have been a perfect example. I updated the post I wrote back in titled "Dow 300 Point Days and Volatility Clustering". An older post here – "More on Volatility Clustering".
Below I take Yahoo DJIA data back to 1929 and the key takeaways are:
1. The market goes up two-thirds of the time.
2. All of the stock market return occurs when the market is already uptrending.
3. The volatility is 80% higher when the market is declining.
4. Roughly 75% of all of the best AND worst days occur when the market is already declining. Reason: see #3.
5. The reason markets are more volatile when declining is because investors use a different part of their brain when losing money. Reminds me of the behavioral reasons listed by Lo in this earlier video.
.jpg)
Below is a chart from my book with monthly data on a few other asset classes.
The title of the post, “Where the Black Swans Lie”, made me think first: black swans do not lie, they reveal unrecognized truth!
i like the new title better . . .
Mebane – LOL about the “missing the 10 best days” and all the variants on that theme! I too think it is a rediculous and misleading stat. For a presentation for a fund idea I have, I ran the counter stat. First, I establised the baseline: $100 invested on Jan 1 1950 (incl dividends) would be worth $5,422 at the end of 2008. BUT, as their argument goes, you missed the 3 BEST days each year, you would only have $52.37!! Sounds like proof positive to buy and hold, correct? Well, if you happened to be so prescient and in fact missed the 3 WORST days?? You'd have $671,573. Equally senseless, useless, and (as you say) misleading figures.
Meb,
I agree as well with your outrage… Mindless drivel for buy and “mold” portfolio management. Missing the best days, which typically often follow the worst days, is difficult if not impossible. Then, of course, missing the worst days BEFORE they occur is as impossible. Funny, you mention black swans, since the 3 best or worst days in a trading year are, of course, tails in themselves… As you also suggest, it is the trend of days in the middle of the distribution that matters more… Bear markets have sharp but brief uptrends, bull markets have sharp but brief downtrends. The profit in either is the more common trend and not the countertrends. With the 10 best or worst days impossible to foresee or avoid, depending on your penchant to go long or short.
What your basically describing is regime switching, whereby you are using the 200 day MA as the identifier of the regime. Have you looked into more sophisticated regime switching models such as Hamilton's Markov techniques?
Outstanding stuff, as usual.
Thank you.
i have not – links? papers?
Very helpful post. A lot volatility assumptions assume price skew but not a vega skew. The comment on regime shifting is pretty helpful. A lot of “simple” models are helped significantly with a regime shift on top.
That new financial database you are getting should be pretty cool. Looking forward to posts with lots of historical and new asset presentations.
Looking at volatility prior to Securities Act of 1933 is unproductive. The margins allowed then contributed to volatility we likely won't see again, and your points will probably be more strongly stated if you rightly exclude the data.
Looking at volatility prior to Securities Act of 1933 is unproductive. The margins allowed then contributed to volatility we likely won't see again, and your points will probably be more strongly stated if you rightly exclude the data.