Below is a table showing the result of buying any time the S&P closes over 1% below its high for the day but still positive by at least 1%:
Between 5 an 9 days out you’ll notice some strongly bullish results. Not visible in the above table is that 19 of 24 instances (79%) posted a close higher than the trigger day within 3 days. Looking out 6 days that number increases to 23 of 24 instances (96%).
8 comments:
sorry to be nitpicky....but can the data's conclusion be due to a lot of occurrences during the 95-00 bull run? ...it's intuitive that during bull markets the market is like a floating ball that won't sink.
an interesting post as always.
cheers.
Anon,
No. Only 2 occurences between 95 and 00. Meanwhile yesterday was the 6th occurence of 08-09 (all since October).
Rob
Anon (above) raises a good point and makes me wonder...just how statistically significant the results are when there is such a small sample size. There are roughly only 20 trades for each day out in your test. If you were to test back decades at least(if you have the data)to increase the sample, would the apparent edge be so great?
The study is already run back to 1978. Prior to that I do not have intraday data.
Rob
This is in response to ANON..
--There are edges that are statistical, and there are configurations that are descriptive or 'diagnostic'.
Descriptics are 'tells' that the market is sending. They enable us to describe w some level of objectivity what point in an Intermediate Move we most likely are at.
Small sample size is not a bar or even a significant factor. How many miracle stories do you need to hear to know that a Saint is genuine. It is a PATTERN cognizing indicator. It is glimpsing a recurring Phenomenon.
Over-bought/Over-sold indicators on the other hand are the opposite. They are statistical and one can NEVER have too many sample datapoints in constructing them. What's the news here, that the market periodically fluctuates to one extreme or the other..?
Clearly the ART in those is to balance several variables into a set point triggering reversal. 43.8% is what the computer spits out when optimizing for one environment. 61.36% is optimal for a different environment.
Etc. etc.
And it becomes personal, and personal style, how much SIGNAL one is willing to miss in order to filter out NOISE. (Which is why Rob is not just blowing smoke when he says he can't predict--or be responsible for--how people use his trading-setpoint ideas).
My own personal preference is for fewer transactions with higher win ratios and higher gains per trade. Partly it’s because I'm a Trustee managing a Retirement Trust and not a professional in the field. Time is a factor for me, as well as P/L concepts. Other subscribers might like his shorter-term, more profitable time-frames.
Here’s what I wrote to my best friend before the Open this morning, based upon my interpretation of the Q-Post in question,
--------
Amigo, CHECK OUT this statement from Q (-ie Rob Hanna’s 4-1 Blog Posting), concerning the end of the First Quarter: “Finishes like Tuesday’s often feel bearish to many traders.... feel inability of the market to hold on to its gains as a potential negative. In actuality...... next 1-2 weeks implications appear bullish....”
Now CHECK OUT this statement from Newsletter Guru Sy Harding, from his daily blog: “On the surface it was a positive day, but that late-day give back was not a good sign.
A perfectly exact example of Rob’s point of a trader exhibiting “feel bearish..potential negative”
This means ultrashort sentiment is on the BULL’s side. Plus there’s always alot of end of Quarter posturing for the Quarterly snapshot, and then undoing-of-posturing, in the last two days before, and 1st two days into, the new Quarter.
In light of above my interpretation of the Q-Post is that this is a Descriptic readout, a probable confirmation of one late-stage intermediate move more (at least) left in the Uptrend. A PATTERN which derives its general success from being a SIGN of bubbly UP action-- but in a healthy 3 forward 2 back kind of progression.
That is, our “Intermediate Bull” move is alive and well, and still capable of snorting out smokin puffs of +2%-- whether that point of expansion comes at 3:59pm EST, or some other time in the trading day. The end-day pullback is indicative of intraday skepticism, a good thing to see in a now stubbornly-resilient Uptrend. (If the final wiggle is meaningful at all-- and not just the ‘noise’ of the abovementioned 2day band of posturing and Daytrader gaming-of-the-posturing). Also, with stubbornly high VIX readings in the background, and longer term newsletter sentiment still quite bearish and dubious, it is hard to see how one can doubt and short such a setup as Q posed...
The scariest BAR would be a panic buying or BUY-fear-capitulation bar rising to the roof and to the last tick of the day, after such a long run pricewise. That might be a painting of a move-ender.
Hope this is helpful, ANON..
Daniel
IMO, you need to do the following in all your studies:
In addition to reporting %wins, avgwin, etc. for each bin, show the equity curve in a small image. It will immediately show whether this is dominated by a certain time period that dominates the conclusions or not, and just give an all around better sense of the data.
DataWolf, an equity curve would be expected if Rob was posting a mechanical trading system.
As you note, it is only a study.
The problem with generating an equity curve from these results is that the exit criteria would be time-based, which IMO is too easily curve-fit to produce anything meaningful.
Well, I wasn't suggesting you would rely on the "equity curve" (which doesn't incorporate txn costs, very realistic entries, position sizing etc.) to draw overly deep conclusions, but just as a quick and dirty illustration of what time periods dominate your returns. For example, observing an inverted u-shaped would be important to note, and can be done without poring through the trades, etc.
It is partly a matter of how you do your studies and how easy it is for you to generate such visualizations. For me, it takes 0 effort to do that, so I always include it when I perform similar studies in order to quickly spot obvious problems (which I am not suggesting exist here.)
Woodshedder, why do you think that time-based exits are more prone to curve-fitting (again, talking not about a mechanical system, but a "study")? I would assume they should be more robust than more complicated exits, since 1. you are not really "optimizing" parameters in those exits and 2. you are making fuller use of your data (for example, if your exits are some kind of profit target based on current volatility, etc. etc. you are in effect only looking whether there spikes/dips of a certain magnitude, and obviously
that would reduce the effective size of the data you are using.)
Post a Comment