Tuesday, December 11, 2012

'tis the season(ality) part 5

here's how i determine whether an equity is on track: i plot a second seasonal projection but anchored on the date ten bars old. here's how i set up the second seasonal projection study - click the edit studies button on the upper tool bar then set the parameters as in the image below:

it is important to set the hpi min pct to some high number to suppress plotting a second hpi indicator which is visually confusing.

afterwards my chart of xrt looks like:

so from this i see that xrt has been closely tracking its season. the season lifts in mid-january and peaks in april.  there is about a 62% chance that xrt rises to 70 or higher by april if the seasonality holds. i plan to be buying xrt in the first week of january, if it executes the dip so much the better.

Tuesday, December 4, 2012

`tis the season(ality) part 4

now let's take a look an equity with some strong seasonality:

 

the mathematical measure of the spread of data is the standard deviation. it was developed to help scientists describe how variable is the repeated measurement of the same thing. in the math literature it is provable that 68% percent of the data points can be shown to fall within one standard deviation from the average. when applying this to seasonality for traders it quickly became apparent to me that what is of interest is when the lower standard deviation line tracks above the flat line in a bullish season or when the upper standard deviation line tracks below the flat line in a bearish season. however, for traders the important thing is how much of the data lies above the lower line and if one uses a 1x standard deviation then 84% of the seasons will track above the lower line. however, if one uses the 1x multiple then one will find that hardly any equity has a season that is so predictable. hardly any of the 1x std lines cross the flat line within a reasonable holding period for traders, say 3 months or so.  the rigorous standard for scientific publication is too rigorous for trading.

in jeffery ma's book, the house advantage,  jeffery claims that his mit blackjack team held only about a 5% advantage over the house when the card count was right. say 52.5% to 47.5%. so i applied this principle to the multiple i apply to the standard deviation of the seasonality. i discovered that 0.3 standard deviations provides a fair but selective number of short term seasonality plays. mathematically it can be shown that 24% of the data falls within 0.3 standard deviations of the mean, or that 62% of the data tracks higher than the lower 0.3x band. if that 62% translates well into the probability of success in a trade then that is a reasonable sort of edge.  to my eye this multiple produces a band that tracks closer to the projection average and provides a much better feel for the seasonality. look at how long the splash hugs the projection average for xrt in the spring before spreading out. to me that's a really good indication that xrt has a tradable season.

however, with this loose a standard, it would not be surprising at all to see that maybe xrt is not on track to repeat this seasonality this time around the wheel. how can we evaluate if xrt is on track with the season? that will be the subject of the next blog ...

Monday, December 3, 2012

'tis the season(ality) part 3

step 3) add a seasonal projection anchored on the most recent candle:
click the beaker icon on the top tool bar.
 
add sdi_seapro5, adjust season len to 5, adjust the colors of the plots fl (flat line) and pa (projection average) to something that contrasts well with your background.

you should now have a screen that looks like:
spy with sdi_seapro5 on weekly chart
this shows you the average move that spy, the s&p 500 etf, has made around this time of year. it looks like we should decline a little til the end of february and then rally into april.

do you trust that?  do you trust it enough to put money on it?

this is the problem i attempt to address with some of the indicators. you can average any random set of prices together but average randomness is still random. to be reliable, the individual seasonal projections should bunch up and move together. here's a view of the individual seasons seapro is averaging (you can plot these by selecting show individual seasons):

spy with individual seasonal projections, green are most recent 2, red are oldest 3.
 
i see the last 2 years were sideways to up, while the oldest 3 years were sideways to down. this is really rather inclusive, imho, and i would not base a trade on this. there are better, more mathematical ways to judge good seasonality which i will explore in the next part.



Sunday, December 2, 2012

'tis the season(ality) part 2

step 2: make 52 bars of right expansion.

go to style/settings/time axis and set the expansion to 52 as in the picture:

one can select more but the seasonal projection will stop after one season of time. selecting less will hide part of the projection but why? the magnification controls on the thinkorswim charts are superb. furthermore, you can select keep time zoom as i have in this dialogue, which will retain the zoom when you switch equities.

Saturday, December 1, 2012

'tis the season(ality) part 1

of all the studies i have created, the seasonal projection studies are the ones i look at every day. however, it has taken some time to arrive at a setup that has that kind of staying power on my eyeballs. if you use my sdi_seapro5 study (seasonal projection of 5 seasons) then you'll want to see how i think you'll get the most out of it. i think there's more than a blog's worth of info I want to put out so this will be a series.


step 1: use weekly aggregation:

i like weekly charts for the simple reason that the most reliable technical analysis is drawn from the weekly charts. my focus on this time-frame is based on the idea that institutional investors are acquiring or closing positions over weeks and months, not days, hours, minutes, seconds, ticks. it is institutions that are waiting with bags of money at these longer term technicals for the right price.

secondarily, i find that the seasonality averages drawn from weekly charts are more valid. most years have 52 weeks, so averaging what happened in week 7 of the last 5 years has validity. some years have 53 weeks. a 53 week year comes about because the last day of the year falls on sunday, the default start of a week. now, the seasonality in my study is based on a moving anchor week, usually the current week and the comparison is averaging what happened multiples of 52 weeks ago (e.g. -52,-104,-156,-208 and -260 weeks ago.) a 53 week year is not really a problem. to see this, suppose it is week 53 of 2006, the last 53 week year. the comparison is going to look at week 1 of 2006, 05, 04, 03, and 02. but this is what one would want anyway because week 53 really only contains trading data from the first trading days of the new year of 2007 (nothing trades New Year's Eve except maybe some cold sores ;-)

however, i do have some reservations about the validity of seasonality drawn from daily charts. there are all kinds of issues with comparing price movements on a day-on-year-ago-day basis. first of all, what season length do you use? most u.s. equities have about 251 or 2 trading days year, depending on leap years. however, if you want to look at futures or forex, they both have about 260 trading days to the year because they trade on US holidays. secondly, there might be a day-of-week issue. if it is monday and the bar 251 bars ago is a friday, is that really a fair comparison? i'm not sure. also, imho, actionable seasonality is an annual event, more or less, when it is present at all.

in any case, the precision of seasonality is not such that it would identify the ideal day to enter a trade. it can be +/- two or three weeks. later, i will show how i identify good seasonality setting up and how i time the season.

next up more chart settings ...