Seeking Alpha

Karel Ondriash's  Instablog

Karel Ondriash
Send Message
Private investor since 2005. Prior to that, Karel Ondriash managed Software Engineering Department at CNET Channel, currently a division of CBS Interactive Inc. He holds a PhD in Physics and Mathematics from General Physics Institute, Russian Academy of Sciences. Karel Ondriash currently lives... More
  • Trading An Index-Tracking ETF/Inverse ETF Pair May Yield Solid Results  3 comments
    Nov 20, 2012 1:41 AM | about stocks: SPY, SH

    Trading a pair of ETFs - index-tracking ETF and inverse ETF - seems to be very suitable approach for those seeking investment opportunities in every market environment. Indeed, irrespective of what the market's general direction is, an investor may easily participate by simply buying shares of the corresponding ETF. One of the most popular ETF is S&P 500 SPDR (ticker symbol SPY) tracking performance of the S&P 500 Index. It has about $100 billion in assets under management and trades more than 100 million shares on an average day. To compose a trading pare, one may consider, for instance, the first inverse ETF ProShares Short S&P 500 (ticker symbol SH). According to its prospectus, "the Fund seeks daily investment results, before fees and expenses, that correspond to the inverse (-1x) of the daily performance of the Index". Though this ETF is much smaller in size, it still possesses very respectful figures: more than $1.5 billion in assets and several millions shares changing hands every day. Thus, both ETFs provide more then enough liquidity for a vast majority of individual investors.

    It's a common knowledge that proper timing is a key to investing success. However, this can be especially true in a volatile market like we observe nowadays. To produce timing signals for trading the pair of index-tracking/inverse ETFs, a combined methodology (NYSE:CM) has been developed. It involves the following methods:

    • Point&Figure (P&F) analysis of the S&P 500 Index;
    • Using William J. O'Neil's concepts to define market bottoms and market tops by looking for follow through days and analyzing distribution respectively;
    • Applying some classical indicators like Moving Averages, Slow Stochastic Oscillator and Bollinger Bands.

    Each method delivers its specific contribution to the CM. For instance, P&F analysis is known for eliminating insignificant price movements, thus helping to identify the trend. Analyzing distribution often allows spotting market weakness in a timely manner, etc. As both methods still generate false signals from time to time, applying classical indicators in addition to them is beneficial to the overall performance. Of course, this gives just a brief idea of how the CM works; its real implementation is a rather complex rule-based system.

    The combined methodology has been back-tested over the period since January 2008 till mid of November 2012 (i.e. over almost 5 years). During this period the market passed through several very different phases. First, after marking a multi-year high in October 2007, the market started relatively slow drifting downside in the first half year of 2008. This was followed by the Leman-induced market crash in September 2008 and sharp decline in subsequent several months. A turnaround that happened in March 2009 has ignited a new bull market. Though the bulls experienced some difficulties in the summer of 2010 and in the summer of 2011, the market managed to reach new multi-year high in September 2012. Thus, the period under consideration looks to some extent as a model of the market cycle.

    A pair of ETFs consisted of the already mentioned SPY and SH. While calculating possible outcome, timing signals were considered to be generated on the basis of the post-market analysis. If a timing signal arose, a market order to buy/sell shares of the appropriate ETF was simulated before the next day market open. Thus, the shares were always considered to be bought/sold at the market opening price.

    Fig.1 illustrates how the hypothetical $1000 investment could have fared if it has been deployed in January 2008 and then was managed consistently according to the timing signals generated by the CM. A cumulated result since January 2008 till mid November 2012 totals about 500%. This corresponds to an average annual growth of approximately 38% during 5 years. No trading fees were being taken into account. However, as the number of trades averages just about 7-9 per year (see Tab.2 below), the profit reduction accounted by trading fees can be treated as being negligible.

    Fig.1 Value of the hypothetical $1000 investment if it was invested according to the proprietary market timing signals since January 2008.

    Each point represents value of the investment after market close.

    (click to enlarge)Fig.1

    Due to the inverse component, the CM delivered strong gains in 2008, when the stock market crashed. In 2009, the CM overtook S&P 500 by a wide margin as the inverse component enabled it to catch the rest of the downside move while participating in the index-tracking ETF ensured benefiting from the market's rebound from its very beginning. In 2010, the CM navigated smoothly through the summer volatility that followed the so-called Flash Crash. A year later, the CM correctly handled both the sharp decline in August and the subsequent strong rally. Current year is probably the most challenging period for the CM so far. Stop-losses have been activated several times recently. This helped to avoid losses, but on the other hand some profits have been missed too. Still, as for November 16, the CM overtakes S&P 500 by about 10 %.

    The following table provides a comparison of the combined methodology performance versus S&P 500.

    (click to enlarge)

    Tab.2 shows some statistics that may be of interest while considering the CM's effectiveness. The number of completed trades averages about 7-9 per year, with the highest number of completed trades - 9 - in 2008 and the lowest number - 7 - in 2009 and 2010.

    (click to enlarge)

    Finally, Tab.3 presents all timing signals generated by the combined methodology during the back-testing period (i.e. since January 2008 till mid of November 2012). BUY signal for the index-tracking ETF sometimes comes simultaneously with SELL signal for the inverse ETF and vice versa. The table gives also an idea on amount of gains/losses in particular trades.

    Tab.3 All timing signals since January 2008 till mid November 2012.

    Conclusion: A combined methodology has been developed that produces market timing signals basing on the post-market analysis of the S&P 500 Index. The methodology was back-tested in application to trading a pair of ETFs - index-tracking ETF and inverse ETF . A cumulated result since January 2008 till mid of November 2012 totals about 500%. This corresponds to an average annual growth of approximately 38% during 5 years.

    Disclosure: I have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

Back To Karel Ondriash's Instablog HomePage »

Instablogs are blogs which are instantly set up and networked within the Seeking Alpha community. Instablog posts are not selected, edited or screened by Seeking Alpha editors, in contrast to contributors' articles.

Comments (3)
Track new comments
  • Hypnos7
    , contributor
    Comments (141) | Send Message
    The performance of your model over the backtesting period is very impressive at 38% average annual gain without leverage. Two questions:


    1) How does the model do prior to 2008? To only test back to 2008 presumes that there was a phase change in the market at that point, and the market going forward will be consistent with this phase.


    2) How many parameters do you have in your model, and how does this compare to the number of trends during your backtesting period? If they are comparable, it is easy to engineer a model which catches all the trends in a given period, but may fail to do so going forward; that is, fall prey to overfitting.


    Thank you for sharing your thoughts.
    26 Nov 2012, 07:51 AM Reply Like
  • Karel Ondriash
    , contributor
    Comments (6) | Send Message
    Author’s reply » Thank you for your questions, Hypnos7.


    The 38% average annual gain is a consequence of the strong market moves in 2008 and 2009. This helped the system to deliver extremely high returns during these years (see Tab.1) thus enhancing the average performance over the whole 5-year period. In my opinion, an annual performance in the range of 25-30% should be more typical.


    Unfortunately, I haven’t back-tested the system’s behavior prior to 2008 yet. I do agree that covering at least one more year - 2007 - would be beneficial for understanding how the system handles the bull market/bear market transition. I’m going to perform this test in the near future. As soon as I have the results, I’ll post them.


    The aggregate model currently consists of 16 independent rules that generate buy/sell signals separately for long and inverse ETFs. Every rule is actually an algorithm dealing with up to 7 parameters. These parameters come from P&F analysis (current market P&F pattern, price objective, and overall market trend), from analysis according to O’Neil’s concepts (current market status), and from 3 technical indicators (statuses of MAs, SSO, and BB). The parameters may have multiple values.


    Each rule has been initially developed to handle a specific market setup that had occurred at least once since January 2008. The market patterns do tend to repeat, however. Every rule has generated more than one signal already. In case of a major market move, the same signal is usually initiated by several rules either simultaneously or within a short time-frame.


    The model has some built-in constraints, including stop losses. So, if a new market pattern emerges from time to time, possible downside is limited. Afterwards, adding a new rule to the system may be considered.


    It seems like this trading system can totally fail just in one case – if the market completely changes its behavior with all the market patterns changing all of a sudden too. In my opinion, such a development is not likely to happen.




    27 Nov 2012, 02:22 PM Reply Like
  • Hypnos7
    , contributor
    Comments (141) | Send Message


    Thank you for your detailed reply. I see the following risk in your approach:


    2008 was a huge trending year, and since then the market action has been heavily influenced by Fed policy. Given the specificity of your market setups (16 rules w/ up to 7 parameters each) even a slight change in the character of the market (e.g., change in Fed policy) could drastically reduce their effectiveness.


    I would encourage you to backtest your model much further back -- there are low-cost ways to obtain daily US market data for several decades. If your model behaves well through several business cycles and Fed policy epochs, then you will have falsified my concern and be more confident that the model will perform in the future (there are no guarantees, of course!).




    27 Nov 2012, 05:18 PM Reply Like
Full index of posts »
Latest Followers


  • If $SPX closes in green today, I will consider adding to my short hedge.
    Apr 9, 2013
  • Bought some $FAS @120.05 #stockaction
    Dec 21, 2012
  • Thursday, Dec 13 - still "Hold long ETFs" status. High odds of a market pullback near-term. $SPY $SH #stockaction #ETF
    Dec 13, 2012
More »

Latest Comments

Posts by Themes
Instablogs are Seeking Alpha's free blogging platform customized for finance, with instant set up and exposure to millions of readers interested in the financial markets. Publish your own instablog in minutes.