 |
| click on images to enlarge |
Rather than another tedious 1929 to the present comparison, here is a less tedious one. I wanted to highlight a couple of points. The first is that the spread between the actual price and predicted price, as measured by the yellow Z-score has now broken the record for historical extremes on the upside.
The second point is made by the red circles on the charts, showing where flat patterns developed during relief rallies just after 1929. The technical term for this is "frog jumping down steep hill" also known as "the toad tilt."
These time series are multifractal, as discovered by Benoit Mandelbrot and others. This means that scaling may occur simultaneously along the time and the price axis. In other words, one pattern may match another by stretching or compressing both time and price scales.
If this model is correct, we should soon be traveling down a steep slope, probably by next week. If prices increase much more, or even stay relatively static for a few trading days, and the Z-score goes to >5, then I'll have to retire this structural equation cycle model. Either way, there is a lot to be learned depending on what happens next week.
I'll be answering some thoughtful posts over the weekend, good luck and good health to all!
Visit econocasts.net for free sample trading model downloads.
9 comments:
Paolo, as you pointed out it looks your model has insensitivity to panic selling. Maybe now we're in a panic buying phase. That's a steep hill that you have the frog jumping down. Do you think the slope cold be causing bigger hops?
Thanks for your interesting work. Can't wait to see what happens next.
Absolutely correct, it is insensitive to both panic buying and selling, and you said it more concisely than I Greenface.
I tried to take that into account during earlier version of the cycle model, and they all failed.
One of the reasons I think they failed is that the daily/weekly short term price series is absolutely not random by any number of statistical measures. In other words, traders clearly have agendas - price level targets that are determined 'a priori' or somewhat independently to the information provided the current price level. (This is where I wish English was my first language!)
That explains why I use pure time series pattern recognition, a type of structured model, to inform shorter term trading. That type of model is very sensitive to "agendas."
This is the first time I have weighed any cycle model so heavily in my trading decisions. I can't wait to see what happens next as well. If the cycle model is farpotshket*, c'est la vie! I still learn something new.
*(Yiddish): Something that is all fouled up, especially as the result of an attempt to fix it. [adjective]
Thanks for prmpting me to attempt to clarify the issue.
Before you "retire" the model assuming there's no correction next week, you might want to take a look at this ZeroHedge article explaining the recent inexplicable runup in stocks:
http://www.zerohedge.com/news/biggest-market-headfake-ever-wholesale-french-bank-liquidity-run-sole-reason-euro-and-sp-surge
--Greg
Greg, I saw that interesting explanation for the anomalous USD/EUR signal with respect to the equity markets. In the case of the DJIA model, which goes back to 1896, other instances of anomalous signals resulting in "panic buying," or "panic selling" for that matter, should have been factored in by the model.
If the model fails, it fails. The unfortunate part is that while the model is adaptive, and adaptive computer runs are relatively short (hours of CPU time) a new model would take at least a month, simply because of the CPU time involved.
Next week should be amusing.
Paolo,
Have you ever run your model across other asset classes? such as some of the currencies and/or other commodities. Is there a certain amount of data which you deem necessary for the model to be relevant.
Monday was not a bad start to your view. Let's see what the rest of the week brings.
Regards
Oliver
Hi Oliver, I guess I am a melancholic, but Monday was a statistical "blip" even if it went in the right direction. Hopefully, the end of this week will reveal some small truths either way.
You can set up this kind of cycle SEM for any time series as long as you remain aware of the assumptions so as not to over-interpret the validity of the model. The most trivial example being to avoid the temptation to add cycles that are much longer than the data set, even if the SEM structure permits it.
The longer and cleaner the time series data set, the better the chance of cycle/fractal discovery. I've spent hours cleaning up financial time series that were supposedly clean. Grumble. Grumble.
Here's the FTSE back to 1800.
FTSE ALL-SHARE 1800
Paolo,
Any update on the latest Z-score after Tuesday?
Regards
Oliver
Hi Oliver,
Yes, here is a running total since last Friday for the DJIA.
Date Actual Pred. |Z|
2011.10.14 11644 9322 2.94
2011.10.17 11397 9186 2.85
2011.10.18 11577 9134 3.13
I'll be happy to update this until the charts on Friday. Thanks for checking in! Paolo
Oliver, I'll put the updates on the Z-score here:
Econocasts business cycle chronicles: DJIA Z-score of cycle model differences
Post a Comment