In fact, Jeffrey Hirsch, author of The Stock Trader's Almanac and reigning king of stock market , argues that when a down Friday and down Monday occur together , it often signals a major stock market turning point-- occasionally a bottom, but more commonly a top:
"For over 30 years, a down Friday followed by down Monday has frequently corresponded to important market inflection points that exhibit a clearly negative bias, often coinciding with market tops and, on a few climactic occasions, such as in October 2002, March 2009 and December 2018, near-major market bottoms. . . . Since 1995, there have been 248 occurrences of Down Friday/Down Monday ( DF / DM ), with 69 falling in the bear market years of 2001, 2002, 2008, 2011 and 2015, producing an average decline of 12.1%." Hirsch, Jeffrey A.. Stock Trader's Almanac 2020 (Almanac Investor Series) . Wiley. Kindle Edition. " (Hirsch, Jeffrey A.. Stock Trader's Almanac 2020 . Wiley. Kindle Edition.)
I've illustrated this dynamic on the chart. Note that there are definitely a few false alarms, but the signal does tend to coincide with inflection points. For instance, most of the significant market downturns in 2019 either began or ended with a down-Friday + down-Monday signal. In March 2020, a down-Friday + down-Monday signal both began and ended the major market plunge associated with the Covid-19 pandemic. And when the market turned downward in September 2020, the down-Friday + down Monday signal twice preceded a bounce.
So, beware, because this last weekend we had a down Friday + down Monday, so a correction may be coming in the near future. (In this case, I think the risk that traders are signaling is related to the possibility that Congressional negotiations for a new federal stimulus package might break down. We also had a major deterioration of jobs numbers last week.) This was a small downward move, and the size of the correction often corresponds to the size of the signal. So, probably don't rush to gamble on a 40% correction or anything crazy like that. But it might be a good time to own protection against, say, a 5% downward move.
I took market data for the last "y" years (where "y" is 3, 5, 7, or 10) and grouped it into a series of sequential (overlapping) pairs of days. Of these pairs, I identified which ones were DFDM.
For every day-pair in the sample, I then computed the change in price over the "n" days (where n = 3, 5, 7, 9, 11, 13, or 15) before and after the day-pair (not including the price change that occurred during the day-pair itself). I took the absolute value of the difference between the post-day-pair change and pre-day-pair change. This gave me a measure of the "magnitude of the inflection" or "change in momentum" that occurred at each day-pair in the sample.
I then subtracted the average magnitude-of-inflection for DFDM day-pairs from the average magnitude-of-inflection for all non-DFDM day-pairs in the sample. This gave me a measure of the "excess inflection" associated with the DFDM signal.
With my 4 different values of "y" and 7 different values of "n," I had a total of 28 model specifications. In every single specification, the average magnitude-of-inflection was higher for DFDM than non-DFDM day-pairs. Effect size ("excess inflection") ranged from $0.213 per share to $3.09 per share, depending on the model specification.
If anyone would like access to the R code, DM me.
3 days | 29%
8 days | 14%
13 days | 14%
18 days | 7%
23 days | 13%
28 days | 9%
As a percentage of the price of the index, these numbers are pretty small. You're getting about 0.6% excess inflection in those first three days as a percentage of index price. So, definitely a little more volatility than usual, but not huge and mostly evident on a short time frame. Perhaps enough to turn a profit on a quick option play.
Moreover, the extensive literature from behavioral economics also challenges the idea that markets are strictly logical. The existence of stock market cycles and seasonality has been known for decades, but their predictive power has not collapsed.
I think DF+DM's track record speaks fairly eloquently for itself.
Are financial markets really scale invariant? Do chickens really lay eggs? Does the Earth really spin on its axis? Let's get to the bottom of this question! And publish a scientific article about it!
But sometimes it is enough to read the abstract/conclusion at the end. And if you do that in case of 9705087.pdf, it basically says "a very interesting question, we can't say anything definitive about it, further research is needed". It doesn't challenge anything, because the data and methods used are just too poor for that.
Do you have a statistical analysis of the DF+DM indicator you'd like to link, or is this just an armchair analysis?
Since I don't know what "an armchair analysis" is supposed to be, I can't link anything, but do you really think a 3% success rate over an average coin flip, is worth it?
Don't get me wrong, I really don't want to be a killjoy, your idea is interesting, I just want to put it in perspective without being too euphoric.
I haven't crunched the numbers on the DF+DM indicator yet, but you've inspired me to do so. I'll get back to you when I'm done.
In regards to scale invariance I think that it does hold semi-strongly but only when markets are efficient and the world is "normal" (when there is sufficient liqudity and we are not in some form of an insolvency crisis). I think assuming that markets are completely scale invariant is a bit wrong and if markets are truly a coin flip, then there wouldn't be any point in trading short term.
If you do code or script
1) You can run different normalized windows of size 10 (2 trading weeks) or whatever over index price data. So for X data points you would have X-10 windows of size 10. If you take the last price in each of the windows and merge it with all the last values of all other windows into one normalized time series signal and try to correlate it with a white noise signal, it is somewhat mean reverting and correlates semi-strongly. So scale-invariance does seem to hold with some datapoints being anomalies like march and sep 2020.
2) Then if you then try to only fit it with down Fridays and Mondays plus the next two weeks it starts to breakdown down and not correlate as much with white noise.
Although when I did this analysis there are less samples in step 2, this anomaly is still useful!