The Dow Jones Industrial Average has a brief downward spike on the 19th of every month since May 2021, except where the 19th falls on a weekend - so, May 19, June 18, July 19, August 19, September 20. But it doesn't seem to do this in earlier months.

My friend speculated that this might be related to options but I'd like to know the whole story. If the downward movement is predictable then shouldn't someone be capitalizing on it?


The Super Bowl Indicator, attributed to Leonard Koppett, states that a if it is won by an original National Football League team (pre AFL/NFL merger) then the market will be up for the year. If the AFL team wins, the market down will be down. In the first 23 years, it was correct 21 times or 91%.

As with your observation about the market dropping on the 19th of the month, this is just a function of randomness. If you look hard enough, you'll find other mundane events that are highly correlated (or highly uncorrelated) with the stock market.

And no, this has nothing to do with options.

  • I wouldn't say that it has nothing to do with options, while it is most likely spurious correlation, coinciding option and futures contracts expiry known as "witching hours" can cause otherwise unusual market activity, which happen to occur around the dates listed here.
    – windwally
    Oct 7 at 16:03
  • The "witching hour" is the last hour of trading on the third Friday of each month when options and futures on stocks and stock indexes expire. Given that the OP's 6 dates occurred on four different days of the week, I'd say that his market drop observation has nothing to do with the "witching hour". And even if all 6 were on a Friday, what's your explanation for option expiration driving the market down consistently ? Oct 7 at 16:45
  • What happens to the market when the Colts win?
    – prl
    Oct 8 at 8:52
  • It's good to know that it's just a coincidence. I was worried that it might have something to do with raising the debt ceiling on the 18th of this month? But I don't even know what that means. Asking people for more money? Oct 11 at 14:46
  • 1
    @Metamorphic - Patterns occur and some people trade them. Note that the indicator was correct 91% of the time during the first 23 years yet during the past 21 years, it was correct 9 times while being wrong 12 times. Google for details about what the debt ceiling is. Oct 11 at 15:33

My guess would be that this is simply a coincidence. Unless someone can point to an event which is scheduled for the 19th of every month and that could affect the stock market. Like if some government agency releases its economic projections on the 19th of each month or some such. If someone on here knows of such a regularly schedule event, please post.

Otherwise, as I say, I'd guess it's just a coincidence. If you look hard enough, you can often find patterns that don't mean a thing. For example, these days whenever I log in to my computer at work I get a six-digit code on my cell phone that I have to type in. I'm always seeing patterns in these numbers. Like, look, today there are three 9's, or the digits are all in ascending order, or there's all even, or whatever.

Sometimes people arguing for some conspiracy theory or mystical interpretation of events will point to such a pattern and then calculate the odds against that pattern. Like say that today my code number has three 9's. If the digits are all random, what's the probability that three or more of them will be 9's? Presumably 1 in a 1000. That's astounding odds! Surely there must be something going on that such a number would come up!

But no, because I'm calculating the probability that this one particular pattern will occur. But you could say similar things if there were three 8s, or if all the digits were even, or hundreds, maybe thousands, of other possible patterns. It's not legitimate to calculate the probability that this one particular pattern will happen. To be fair you'd have to calculate the probability that ANYTHING that you might notice and call a pattern would happen. And of course if you try hard enough, you could find some sort of pattern in any random number.

So is it significant that for several months in a row the stock market dropped on the 19th? Probably not. I'm sure that if you looked at other periods you would find spans of a few months where the stock market dropped on the 22nd, or on one of your friend's birthdays, or on a day when it rained in Kazakhstan, etc.

If you see a pattern like this, it makes sense to look for a reasonable cause. Can you find some event that regularly happens on that day that might plausibly affect the stock market? If so, you may have found a useful tool for your trading. If not, of course it's possible that something really is happenning and you just can't identify. See if the pattern continues. Keep looking for causes.

And sadly but realistically, if there WAS some simple cause, odds are that the professional traders would have noticed it long before you did.


I don't have a full answer to my question, but I attempted to analyze some historical data and compute some statistics and p-values recently. This post is a slight edit of a closed post that I wrote yesterday on the Cross Validated (statistics) stack exchange. The code below is in R.

The Super Bowl Indicator, which was said to be correct 21 times in the first 23 years, has a p-value of 3e-5:

> n=23; k=21; a=0; for(i in k:n) { a <<- a+choose(n,i) }; a/2^n
[1] 3.302097e-05

The Super Bowl Indicator is still significant today, but less so (I disagree with Wikipedia that this is spurious):

> n=53; k=40; a=0; for(i in k:n) { a <<- a+choose(n,i) }; a/2^n
[1] 0.0001342701

Since posting my question I decided to run some experiments to estimate the actual significance of the pattern I observed. I chose to do this by looking at the times when a 5-month series of dips in the Dow has almost occurred in the past, as well as by generating synthetic data according to the modeling assumption that day-to-day market variations are independent from each other.

Historical approach

According to the first (historical) approach, we find that this has never occurred before. Here is a chart of the market behavior in the past 6-month period:

Dow prices, April 2021 - October 2021

The black line shows the closing price of the Dow Jones Industrial Average and the vertical gray lines mark the 19th of the month (or closest weekday) on the months in question. You can see that the largest dips coincide with these gray lines.

Here is the same chart expanded to the full period of Dow prices that I have access to; it is from 1992, the earliest date offered to me by Yahoo Finance:

Dow prices, 1992-2021

The green line shows the detrended closing price data and is obtained by subtracting a five-day average, visible in blue on the close-ups. The pink line graphs a statistic representing the extent to which one of the days of a given month is the start of a sequence of five dips occurring on the same day in the succeeding months. The thick red lines show the value of this statistic for the three most significant (lowest value) 5-month periods. These periods appear highlighted in gray.

Runners up (no pattern)

A close-up of the second and third most significant periods shows that the fit to the 5-month pattern is not as good as with the present period. There are dips missing and they are not as sharp. My statistic ranks those periods in second and third place to the present, and it is easy to see that they do not fit the pattern. This shows that what we are seeing now has never happened before.

Dow prices, August 2010 - March 2011 Dow prices, July 2014 - March 2015

Estimating p-values from synthetic price data

I unfortunately have no plots for the numerical experiment with synthetic prices, which I will proceed to describe. The method I used for this experiment was to select the most recent 6-month period in the data. I then generated synthetic price traces with the same starting and ending values by computing successive differences in the detrended (green line) historical prices for this period and taking the cumulative sum. The significance of the dips on the day in question is then calculated using the earlier statistic. This is done 100,000 times. None of these 100,000 synthetic price traces are found to contain a stronger pattern of dips on the 19th.

In my Cross Validated post I quoted a p-value based on a normality assumption for the distribution of the statistic on the synthetic prices. I said this was 1e-7, but I think that was in error, as I realized my code had neglected to detrend the synthetic traces. The mathematical assumptions are probably still dubious, but the new estimate is much smaller, 0.001:

> mean(pval_res$ss)
[1] 48.34911
> sd(pval_res$ss)
[1] 14.52588
> pval_res$s0
[1] 4.384212
> pnorm(pval_res$s0,mean=mean(pval_res$ss),sd=sd(pval_res$ss))
[1] 0.001236365

Yet the normality assumption must be broken, and the actual p-value must be less than around 1 in 100,000, because the sampling experiment failed to turn up a better price trace. At least I think this is implied, my statistics is a bit weak.

> sum(pval_res$ss<pval_res$s0)
[1] 0
> length(pval_res$ss)
[1] 100000

Either number should be multiplied by the number of days in the data set, 7500. If the p-value is 7500*1e-5, then this is just over 1 in 13, not very significant.

And the fact that the Dow was "normal" yesterday suggests that the pattern has ended.

  • No amount of statistical analysis of the past is going to reveal what may happen in the future. Oct 20 at 18:35
  • 2
    Except it wasn't like somebody developed a theory that the stock market should be correlated with AFL/NFL outcomes or the 19th of the month and then calculated a p-value for the results of their model. Rather, people went looking for weird correlations, and now you are calculating a p-value for them. This is some times called data dredging. You can certainly calculate a p-value, but it doesn't mean it's credible. xkcd.com/882 Oct 20 at 19:41
  • @CharlesEGrant: Yes, model selection has to account for model complexity. And we often rank models by proposing a real-world mechanism for the observed correlations, something that is hard to do in the case of the Super Bowl Indicator. Oct 21 at 7:02
  • If you threw a handful of beans in the air and they landed to form the shape of the word "RUN", you might be confused about the mechanism, but I don't think you'd be confused about the intended recipient of the message. You wouldn't for example infer that this low-probability event was intended to happen in someone else's kitchen, simply because you don't understand it. And you wouldn't probably trouble too much about calculating p-values, or proposing a mechanism, since the hypothesis itself - "someone is communicating the English language verb 'run'" - can be stated quite briefly. Oct 21 at 7:03
  • 2
    Oh my, I think that I have been DOSsed! I look forward to your statistical ability providing anything useful about what will happen in the future with any stock, ETF, market, etc. rather than looking in the rear view mirror and offering 20-20 hindsight analysis. Oct 21 at 13:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.