How does the prior years return impact future performance

In the following study I decided to analyze the walk forward performance of some highly liquid ETFs.  For this study the ETFs are divided into four buckets of performance over the prior year.   The buckets are as follows,  bucket1 (-50% to -25%),  bucket2 (-25% to 0%), bucket3 (0% to 25%),  bucket4 (25% to 50%).  The walk forward periods are 90, 120 and 252 days.   What is interesting is that the best walk-forward performance occurs when the prior year performance is in bucket1 (-50% to -25%). 

The 252 days walk-forward period performance is on average between 45% and 70% if the prior year was in bucket1.   Similarly in all cases if the average over the prior year was in bucket 4 (25% to 50%) in the next 90, 120, 252 walk forward periods the performance is negative.    The trading strategy would be load up  when the prior year is in bucket 1and short when the prior year is in bucket 4 .  Note also the the 252 day walk forward period is significantly better than both the 90 and 120 day periods.

Results For the Brazil EWZ Model

The last long EWZ trade was entered on  08/27/2010 and exited  10/05/2010 right now the system is neutral.  The system had a return of 16% for 2010 vs a gain of 7.07% for the underlyer.  So far the underlyer EWZ has had a net loss since year end 2010 while the system has remained neutral.

Emerging Markets Model Performance Through Jan 2011

The table below summarizes the results of our strategy as applied to the EEMs (Emerging Markets ETF).  We exited a long position in late November 2011 and are currently neutral.

Cumulative Performance Of QQQQ Model

In this post I want to summarize the performance of our QQQQ model from inception.  We recently exited a long position on January 14 2011 and are currently neutral.   The QQQQ model is an example of our slow trading systems which identify longer dated trends and stick with them.   The key to the model is to avoid major pitfalls and downturns.   The model can be either in or out of the market for long periods of time.  Most notably when the NASDAQ bubble burst in 2000 the model stayed out of the market while the NASDAQ lost half its value.   The model goes long in 2002 and avoids the meltdown in 2008 going long again in 2009.   The purpose of this model is identify long term trends.   It is worthwhile noting that if you invested $1000,000 in the QQQQ at the end of April 2000 you would still be down over 40% by today.  Whereas if you traded the QQQQ ETF using the model you end up 340% over the same period with much less volatility.

A Correlation Study between ETFs

The other day I was reading an article in Seeking Alpha by Erik Gholtoghian in which he established a multi-factor capm model between DRYS, SEA and USO which can be summarized as follows:

 DRYS weekly % change = 1.38*weekly% change in SEA+.56*weekly %change in USO.

SEA is the Guggenheim/Delta Global Shipping ETF which was introduced in the middle of last year, OIH is the Merrill Lynch Oil Service Holders ETF.  I decided to conduct a  study that examined the correlation of SEA with respect to commodities that are shipped by DRYS container ships.  In the first chart the correlation matrix includes data from the SEA ETF which started trading in June last year.   The strongest relationship exists between XLE and IYM for  06/14/2010 till present and from jan 3 2007-present.  There is also a strong correlation between SEA and IYM and XLE over the much shorter period.  These represent potential trading opportunities assuming that pairs trade within certain ranges and they mean revert.

The first chart below summarizes the relationship between IYM and XLE from January 2007 till the present.  Clearly there are buying opportunites when IYM trades below XLE and selling opportunties when IYM crosses above XLE.  There are a few ways to play this one could go long IYM at some point after crossing XLE from above and waiting till it once more crosses XLE from below.  A second trade is to buy XLE and Sell IYM when the spread between the two has widened and close out the position when it converges.

In the second chart below I incorporate the SEA data from 06/14/2010

till present. 

Some Interesting Charts

I decided to perform a study which examined the relationship between DRYS which is the largest carrier of dry bulk goods by market cap vs some key ETFs.   DryShips, Inc., engages in the ownership and operation of drybulk carriers and drilling rigs that operate worldwide. Its drybulk fleet principally carries drybulk commodities, including coal, iron ore, and grains; and minor bulk items, such as bauxite, phosphate, fertilizers, and steel products.  DryShips also engages in the shipment of oil based products throughout the world.   The ETF's I selected for the study were as follows  IYM (Basic Materials),  SLX (Steel),  XME (Metals and Mining),  OIH (Oil).  All of these ETF's have exposure to stock holdings in the various sectors.

In the first chart below I divide the ETF Close Price by the DRYS Close Price for that day.  These ratios achieved their low points in October29 2007 this coincided with a peak in DRYS.  Note that IYM, SLX, XME and OIH peaked in August 2008 so that the peak in DRYS preceded the subsequent collapse in IYM, SLX, XME and OIH by ten months.   The various ratios all achieved another peak in early March 09 which represented a major buying point in the markets.   These ratios are once more achieving new highs as we speak which bodes well for for IYM, SLX, XME and OIH.  I will analyze this line of thought further for my next posting when I will examine both the current and lagged returns of DRYS vs the ETF returns.

For Gold And Silver Bugs

Below I am enclosing the results of my gold and silver models.   The trend following models I employ are eant to capture long established trends in the underlying assets.   The GLD model entered the position in April 2005 and remained long through mid October 2010.   Note that in this particular instance the model more or less mimics the underlying GLD ETF. 

 The Silver Model modestly outperforms the SLV ETF but does so with approx 40% less volatility.  The rules for entry and exit are established in order to determine exit and entry points.   This methodology has worked well in both the QQQQs, EWZ, EWH in particular (refer to my first posting for results through Nov 2010).
The idea is to have a model that captures long term trends in multiple markets by applying a similar process for determining rules.   I have been playing around with some shorter dated models and will post the results of those when I am satisfied with the progress.   The ideal end user of these models would be fund managers adopting a long term horizon who may either use this model to trade these ETFs or to use it as an overlay to manage there own stock picking.  These models will not appy to the HFT community many of whose models success relies on picking off ones customers through the ability to co-locate their servers with the exchanges.