Daily Commentary - Posted on Tuesday, December 22, 2009, 9:42 PM GMT +1

33 Comments


Dec Tuesday 22

How To Make A Million (%) Trading The SPYDER – Part I

Due to the fact that I’m very busy these days building and moving into a new home in a couple of weeks, I spent the majority of the time left for trading and blogging (sorry about the latter) with working on a market model (a concise mathematical formula which would stand the test of time forecasting the next sessions S&P 500‘ performance on the close with maximum accuracy), trying to make gread and fear as the never changing driving forces behind maket movements somewhat ‘quantifiable’.

I started the process of developing such a ‘market model‘ (financial trading strategy) a couple of month ago, but up to now never had the necessary desire and time (unfortunately developing a reliable trading system is a complex venture) to make any substantial progress. But being booked to capacity by serveral personal and business matters, and additionally being inspired by Michael Stokes’ posting Wasting a Good Life Trading, I thought it would be the right time to resume working on the model.

So this will be the first in a series of posts about my my step-by-step approach, initial basic parameters, and respective results.

In order to keep things as simple as possible right at the start (leaving room for further improvements), my initial basic parameters were:

  • For backtesting the SPY (SPDR S&P 500 ETF) will be utilized (adjusted for dividend and cash payments in order to track the S&P 500 as close as possible) which corresponds generally to the price and yield performance, before fees and expenses, of the S&P 500 Index. 
  • Positions will be entered into or an open position closed at the market’s regular close only (market-on-close orders).
  • No (intraday) stops (buy and/or sell stops) will be used even if the SPY (S&P 500 EFF) is being utilized.
  • No position sizing, the model is always ‘all in‘ (e.g. no Kelly, optimal f, fixed fraction, …).
  • No leverage taken (no double or triple-leveraged ETFs are used).
  • No abnormal market filter will be used (e.g. during phases of extremly high/low volatility, strong trending markets, the market’s lacking compliance to the model’s forecasts with a resultant number of consecutive losses and/or serious drawdown, and and and)
  • No adaptations (no changes and/or cancellations/additions of formulas, conditions, and model parameters over the course of the lifetime of the model/during backtesting).
  • The model (and respective performance figures) does not account for slippage, transactions costs (commissions, exchange and regulatory fees), and interest on idle balances.

xx

In a first step the model is simply taking a long or short position (never being market neutral, means if no buy setup is triggered, a short position will be taken) on the close (e.g. buying/selling short the SPY), and the model is always ‘all in‘ (as already pointed out no position sizing and/or leverage). Using buy/sell stops, position sizing, adaptations, abnormal market filters, optimization of short positions (no buy setup trigged does not necessarily mean that a short position has to be taken, market neutral may be a better decision if no edge is provided) may be subject to a future optimization process.

Assessment criteria (in absolute terms and in comparison to the general market) for the market model and the selection of setups, conditions and the set of parameters will be (in the order of precedence):

  1. Cumulative Returns and CAGR (Compounded Annual Growth Rate = geometric mean growth rate on an annualized basis)
  2. Growth Rate per Trade (geometric mean growth rate on a ‘per trade‘ basis, means a trades’s average contribution to cumulative profits / geometric growth)
  3. Maximum Drawdown (the maximum decline from a historical peak in cumulative profits)
  4. Sharpe Ratio (excess returns per unit of risk of a financial trading strategy)

The model’s setups and conditions can in principle be grouped into three categories:

  1. Seasonalities
  2. Extreme Market Conditions
  3. Regular Market Conditions

xx

To cut a long story short: The stats and figures belwo represent my current status quo, I’m still working on the model trying to streamline conditions, the set of parameters and their respective dependencies and will report about my progress and my step-by-step approach over the course of the next couple of weeks. But already at this stage it is interesting to note that even without adaptations, market filters, positions sizing, stops and and and it would’ve been possible

  • to achieve positive returns (not a single losing year) in every year since 1990 (I don’t have breadth and other -except index- data before 1990),
  • to achieve returns in excess of +50% in 10 out of the last 20 years, and returns in excess of +100% in 4 out of the last 20 years,
  • to achieve an compounded annual growth rate of 65.14% over the course of the last 20 years,
  • to out-perform the index (regularly by a very wide margin) in 18 out of the last 20 years,
  • to face a maximum drawdown of 16.05% only during the last 20 years,
  • that one would’ve never been in a drawdown more than 118 trading days (6 month).

Listed below are some details with respect to the strategy’s setups, conditions, parameters and dependencies (>indicator x< is a placeholder for a proprietary indicator).

1. Seasonalities

(1.1) it is the session preceding Memorial Day, Labor Day, Thanksgiving Day, or Christmas Day,
(1.2) it is the last business day of the month,
(1.3) it is the session immediately preceding Jobs Report Friday (regularly the first Friday of a month), and the index closed up,
(1.4) it is the session immediately preceding an FOMC announcement day,
(1.5) it is an FOMC announcement day, and the index did NOT closed higher greater than +0.45%.

2. Extreme Market Conditions

(2.1) the index did NOT close 2 standard deviations above it’s 11-day EMA (Exponential Moving Average),
(2.2) the VIX (CBOE Volatility Index) did NOT close lower less than -15%,
(2.3) the 2-day RSI did NOT close above 94, and the >indicator 1< did NOT close above xxx,
(2.4) the CBOE Equity Put/Call Ratio closed higher at least 45% above it’s simple moving average of the previous two sessions.

2. Regular Market Conditions

(2.1) SPY volume came in at least 55% above the previous session’s volume and S&P 500 Advancing/Declining Issues closed above 0.85 and the ratio of S&P 500 stocks penetrating their previous session’s high vs. those penetrating their previous session’s low closed above 0.50,
(2.2) SPY volume did NOT close 30% above the previous session’s volume and S&P 500 Advancing/Declining Issues did NOT close above 1.30 and the ratio of S&P 500 stocks penetrating their previous session’s high vs. those penetrating their previous session’s low did NOT close above 1.40,
(2.3) the index closed above it’s 19-day EMA (Exponential Moving Average) and

  • the index did NOT post an intraday high less than -0.30% below the previous session’s high, OR
  • the index did NOT post an intraday low less than -0.50% below the previous session’s low, OR
  • the >indicator 2< did NOT close above yyy,

OR
the index closed above it’s 19-day EMA (Exponential Moving Average) and (the ratio of the 2-day +DI (Wilder’s Directional Movement Indicator) vs. the 2-day -DI closed below 0.65 OR the >indicator 3< close below zzz) and

  • the index did NOT post an intraday high below the previous session’s high, OR
  • the index did NOT post an intraday low less than -0.15% below the previous session’s low.

(2.4) …
(2.5) to be continued …

A long position is taken at the close in the event a buy setup (in extracts listed above) had been triggered, and a short position if no buy setup had been triggered (up to now the strategy is NOT optimized for the short side of the market, sometimes it would be wise to take no position at all if no edge is provided on the long side).

Table I shows the SPY‘s (S&P 500 ETF) performance (cumulative returns) since 01/01/1990. Setup 2 represents the long side (long trades only) of the strategy, setup 3 represents the short side (short trades only) of the strategy, and setup 1 the overall stratgey as a combination of long and short trades.

2009-12-21-SPY-S1

Figure I shows the respective equity curve (setup 2 -longs only- and setup 3 -shorts only-) from 01/01/1990 to 12/31/1999.

2009-12-21-SPY-S2

Figure II shows the equity curve, now including setup 1 as the combination of long and shorts, from 01/01/1990 to 12/31/1999.

2009-12-21-SPY-S6

Figure III shows the respective equity curve (setup 2 -longs only- and setup 3 -shorts only-) from 01/01/2000 to 12/18/2009 (the SPY’s equity curve is the thin black line at and around the 0%-line).

2009-12-21-SPY-S3

Table II shows the SPY‘s (S&P 500 ETF) performance (cumulative returns) since 01/01/2009 (year-to-date). Setup 2 represents the long side (long trades only) of the strategy, setup 3 represents the short side (short trades only) of the strategy, and setup 1 the overall stratgey as a combination of long and short trades.

2009-12-21-SPY-S4

Accordingly figure IV shows the respective year-to-date (01/01/2009 to 12/18/2009) equity curve.

2009-12-21-SPY-S5

Successful trading,
Frank

________________________________

If you might want to be instantly notified about what’s happening in the markets and at TRADING THE ODDS, I encourage you to subscribe to my RSS Feed or Email Feed, and (or) follow me on Twitter.

xx

Disclaimer: No position in the securities mentioned in this post at time of writing.

The information on this site is provided for statistical and informational purposes only. Nothing herein should be interpreted or regarded as personalized investment advice or to state or imply that past results are an indication of future performance. The author of this website is not a licensed financial advisor and will not accept liability for any loss or damage, including without limitation to, any loss of profit, which may arise directly or indirectly from use of or reliance on the content of this website(s). Under no circumstances does this information represent an advice or recommendation to buy, sell or hold any security.

Add to Technorati Favorites

Comments (33)

 

  1. Jason says:

    Man, thanks for the update. This project looks extremely promising, can’t wait to see it as it is finalized.

  2. Shane says:

    Looks great. For the “Seasonalities”, are you saying that you would only look to go long on those days (i.e. that you would be short on every day of the year not included in that list)?

    • Shane,

      thanks, and yes. The model would go long on those pre-holiday sessions and short on all other pre-holiday sessions EXCEPT an extrem market condition setup would’ve been triggered (there’re some dependencies).

      Best,
      Frank

      • Shane says:

        Thanks Frank. But doesn’t that mean that given you’re never neutral, and can only be long on a small number of days, that you should be short most of the time? That doesn’t play out in your performance numbers where you appear to be long/short roughly 50% of the time. Am I misinterpreting? So having a long signal is based on (1) seasonality OR (2) market conditions, not on (1) seasonality AND (2) market conditions? In other words, you are short when you are not on one of the days in (1) or in extreme market conditions (2).

        • Shane,

          as pointed out ‘A long position is taken at the close in the event a buy setup (in extracts listed above) had been triggered’ (Emphasis on ‘in extracts’). The conditions listed in the post comprise approximately 60% of the strategy’s setups and conditions.

          Best,
          Frank

  3. CarlosR says:

    Very interesting work, Frank. I’ll be following further posts with great interest.

    One thing that concerns me a little is the presence of all those fixed numbers in your parameters for determining the market conditions. My preference would be to have those settings be adaptive, as I feel sure they will change over time.

    On the other hand, having them fixed makes development easier, I’m sure. In addition, if you can get good performance with them fixed, then making them adaptive should only increase performance in the future. As you said in the introduction, this was done to keep things simple and leave room for further improvements, and as a matter of development process, I completely agree with that.

    I’ll be looking forward to further posts in this series, as well as also hopefully the resumption of your other ‘normal’ posts at some time in the future, if time allows. But I know you must be extremely busy these days, and I’m sure you have higher priorities on your agenda, all of which is completely understandable.

    Hope the move to the new house goes well, and have a great holiday season!

    • CarlosR,

      thanks a lot.

      But from my perspective making the trading system adaptive is curve-fitting with hindsight. You only know with hindsight which set of parameters (and conditions) would’ve worked best in the (recent) past, but you will never know in advance when (and to what extend) parameters should be changed (and/or conditions being added, adapted or being completely deleted from the system), the market will not ring a bell.

      Wouldn’t you prefer knowing that the system worked without any adaptations, so there would never be any uncertainess when -and to what extent- to change conditions, parapemeters and dependencies ?

      Setting up a trading system is of course nothing else than curve-fitting with hindsight as a one-time process (only). If you add an unlimited number of conditions and parameters to the system and/or change their numbers and dependencies again and again, you might be able to track the SPY hundred per cent (in the past). I’m always carefull of those system which claim a high winning rate or profit factor, but change conditions and parameters at frequent intervals.

      From my perspective the most interesting part is that the system worked without any adaptations, especially the long only part of the system would’ve achieved extraordinary results and stood the test of time at least for the last 20 years.

      Best
      Frank

      • J Marc says:

        That last Paragraph, “no adaptions” very good, Seems robust . Others that keep adapting while they has their arguments also shows that at any time are you going to have to tinker? I prefer your thesis, if it is working. Seems better.
        Thanks . Looks good.
        Joe Marc

      • mhf says:

        It appears that the numbers (parameters) you chose for most of your indicators look like they “soft of work” when plotted on a chart. For example, why did you choose 94 for RSI(2) in 2.3 and not 95? Ask any psychologist and he/she would likely tell you that humans are biased to picking a round number like 95. So what’s your reason for going 1 less? To me that smells precisely like curve fitting.

        You call adaptation “curve-fitting with hindsight” and you say that “you will never know in advance when (and to what extend) parameters should be changed”. True. The hypothesis behind adaptation is that the markets change slowly, but once changed they have a momentum (the new parameters work until they change again). I can definitely see the logic behind rejecting adaptation, but with that said, are you implying that you will be using the same exact parameters given in your post for the next 50 years? If not, how will you know when you change them? If yes, how would you account for change in next day momentum pattern circa 1998?

        • mhf,

          I choose 94 for RSI(2) because it worked best during at least the last 20 years. The difference between 94 and 95 are approximately 120,000% in cumulative returns. “Ask any psychologist and he/she would likely tell you that humans are biased to picking a round number like 95″: That is exactly the reason why most people loose money trading the markets, because they prefer picking round numbers instead of using what worked best (thinking they ‘should’ use a round number), or making one or more of those countless and costly mistakes. Why should I use 95 for RSI(2) leaving a lot of money on the table when 94 would’ve achieved far superior returns ? Please explain why using 95 will not be defined as “curve-fitting” while 94 will ???? Because most people would be biased to choose 95 ? I’m happy that I’m not ‘most people’, because trading the markets it never pays to be part of the masses or being one of those ‘most people’.

          I do not reject adaptation, but would prefer a system which is robust enough to provide outstanding returns without the necessity to make (frequent) changes/additions/deletions to conditions/setups/parameters.

          Best,
          Frank

          • mhf says:

            My apologies if my comment angered you. My objective was not to pick a fight. Group hug :)

            Regardless, I simply want to see your justification for not using out-of-sample or walk-forward testing. I believe your justification for this (please correct me if I’m wrong here) is that the parameters you’ve picked make the system “robust enough to provide outstanding returns”. But how do you define “robust” trading system? I guess my main concern is how do you justify the stability of the parameters you’ve selected? Commonly this is done by comparing in-sample performance (which is what you have here) and out-of-sample performance (which you didn’t provide).

            Thanks,
            mhf

          • mhf,

            no problem, and it wasn’t my intention to offend your sensibilities as well.

            With respect to out-of-sample or walk-forward testing, the odds of selecting something which was all by chance (and purely random) decreases exponentially as one uses a larger test set (in this case more than 5,000 data points/trading days). In addition, I’d have come up with the same set of conditions and parameters using the in-sample data period between 01/01/1990 and 12/31/1999 only. Utilizing the t-test and the out-of-sample data period from 01/01/2000 onward for verification purposes (only to make a quit shot, there are better methodologies for verification), the probability that the positive performance occured by chance only is less than 50% in 2006, less than 25% in 2003, and less than 2.5% in all other years (compared to the market’s at-any-time performance). For my purposes that is reason enough to go on with the model …

            Best,
            Frank

      • CarlosR says:

        Frank,

        I do agree with you that a fixed model that worked well going forward could be a very nice thing, indeed. I think one could feel psychologically more confident in it, and as you pointed out, you never would have to worry about when to make changes.

        However, my experience with models like that (mostly developed by others), including some developed using 12 and 15 years worth of data, is that they can become obsolete very quickly. Frankly, I was surprised when I saw that, because it is the opposite of what you’d expect, but there it was.

        I hope your model will be different. Time will tell, as it always does, but I would not be surprised if somewhere down the road you decide to add some adaptation.

        In any case, I wish you the best of luck with it, I’ll be rooting for your success.

  4. joe duffy says:

    The Profit Factor and Win % look good. If you run the same parameters on the SP index as a proxy for SPY ETF back to 1997 how do the results look over that period?

    • Joe,

      thanks.

      1) Forget about the Win %, although it regularly (but undeservingly) gets the most attention. Think about a system being right on 9 out of every 10 sessions making an average profit of 0.10%, and losing on 1 out of every 10 sessions only, average loss -2.0%. Win %: 90%, expectancy -0.11% per trade (90% * 0.1% + 10% * -2.0%). An extraordinary accuracy of forecast, unfortunately this system would shredder your money.

      2) I can’t use the SP because the SP’s opening quotation’s (the same applies to the DOW, NDX, RUT as well) are NOT accurate. Utilizing the SP (with the opening quotation being part of the system’s conditions) the system would significantly differ from utilizing the SPY or the ES E-mini S&P 500 (both tradable assets with accurate opening quotations).

      Best,
      Frank

  5. BGPL says:

    hi Frank,
    nice work !
    A comment: (more for my education).
    Most very short term mean reversion strategies worked very well since 1990…
    I cant (yet) tell based on the rules if indeed it is such a strategy.
    Would be interesting to hear your comments on that -> just from a perspective of my interest in developing non-correlating strategies.
    I find your some of your choices of filters in seasonalities and for extreme market conditions educative – thank you.
    best wishes..
    bgpl

    • BGPL says:

      i forgot to add: one of my suspicions that this is short term mean reverting in nature is that the 2009 ytd performance seems to have levelled off in the last 3 months – similar to some of the short-term MR strategies i am tracking. But it obviously a very poor piece of evidence.. ;)

  6. tito says:

    > I choose 94 for RSI(2) because it worked best during at least the last 20 years. The difference between 94 and 95 are approximately 120,000% in cumulative returns.

    Changing one integral parameter by 1 has this kind of effect? Yowza.

    This clearly indicates that the strategy’s returns are severely discontinuous over the parameters you’re ‘optimizing’ it over. To me this implies that the strategy itself isn’t robust and that the felicitous-looking results you’re looking at are almost certainly due to over-optimization… This breaks my first rule for determining if a strategy is a winner.

    What am I missing?

    • tito,

      120,000% in cumulative returns sounds much, but is nothing more or less than a 5% move (for example on a single session) in your favorite direction (2’1 mio. % * (1+5%)). Do you really think that one (big) winning trade more or less (out of 5,031 trades) would turn a robust trading strategy (if it would be one) into a non-robust strategy ?

      That’s simply the power of compounded returns over more than 5,000 trades. Any small changes to conditions and/or parameters might have (better: most probably will have) a huge impact on compounded returns.

      I’ve no idea why people think if a trading system doesn’t work with ’round numbers’ like 5/10/75/95 for RSI(2) instead of 4.5/9/77/94 (among others), or something like ‘a close < 0' (instead of 'a close < -0.15%)', 'it won't work', 'it's not robust', … Trading the markets it never pays to be part of the masses, to follow what others think one (you) should do, to do what most people are 'biased to do'.

      Best,
      Frank

      • tito says:

        > Any small changes to conditions and/or parameters might have (better: most probably will have) a huge impact on compounded returns.

        That’s not generally true, Frank. In fact, the more stable a strategy’s returns across parameter permutations, the more robust it will prove to be in practice. Strategies that don’t exhibit stability are – in my experience – uniformly losers in practice. It follows that a strategy that performs well on the spy should likely perform similarly on another broad-based equity index. And if you can identify a tradeable anomaly on an equity index, you can essentially always outperform it with a suitable portfolio of equities.

        > Do you really think that one (big) winning trade more or less (out of 5,031 trades) would turn a robust trading strategy (if it would be one) into a non-robust strategy ?

        No, I don’t. If small changes to the strat’s params make big changes in its performance, then it’s not a robust strat in *either* form! ;^>

        > I’ve no idea why people think if a trading system doesn’t work with ’round numbers’ ..

        I didn’t suggest anything of the sort!

        In my orig comment, I’d tried to include a link to a post I’d written on this topic some time back, but it seems this blog doesn’t like my html-enriched comments… I’d written that the key elements that identify a strategy as a winner and not just a pretender are:

        1. it’s not excessively sensitive to particular parameters
        2. even the worst permutations of the strategy are winners (they beat relevant benchmarks by volatility-adjusted measures)
        3. it’s successful across a variety of time periods
        4. it’s successful for long time periods
        5. it’s successful across a wide set of markets

        This strategy, as described, fails #s 1, 2 and 5 which is a huge warning sign to me…

        Anyway, it’s very nice of you to share your process and I assure you that I hope you are right and about to make a supreme fortune with this strategy. But I’ve spent years looking at the simulated and actual results of trading strategies and this one looks fishy…

        ps – your English is better than that of most Americans – Bravo!

        • tito,

          thanks a lot for your kind words (especially with respect to my English).

          With respect to ‘This strategy, as described, fails #s 1, 2 and 5 which is a huge warning sign to me…': I never mentioned that the strategy would be excessively sensitive and dependend to particular parameters (although the power of compounded returns makes a huge difference), that a profitable strategy would turn into a losing one with the worst permutations (in fact it won’t), and that it wouldn’t be successful on other broad-based equity indices (it would’ve been profitable on all major US major market indices, although to a significantly lesser extent, mainly due to the fact that the opening quotations for the INDU, NDX, RUT and SPX are not acccurate).

          But I’m not looking for the Swiss Army knife of trading strategies, therefore I do not agree to all your bullet points 1 to 5 (although I know that the majority of those bullets points are regularly cited in different books about trading). To 3. and 4. I fully agree, to 1. I agree in parts, and I personally do not agree to 2. and 5.

          From my perspective the objective of bulletpoint 2. is regularly to increase one’s confidence that you won’t loose your shirt if things won’t work expectedly, but there are better measurements to increase the confidence level (Sharpe Ratio, Maximum Drawdown, Time in Drawdown, and so on).

          Think about a strategy buying the SPY on close of an pre-FOMC announcement session and on an FOMC announcement session if the SPY didn’t close higher than +0.45%. This would be a highly profitable strategy over the course of at least the last 20 years, it would’ve worked for all major market indices, it would’ve worked across a wide variety of time periods, but it would’ve been extremely sensitive to it’s parameters (pre-FOMC and FOMC announcement session, and a maximum performance of +0.45% on an FOMC announcement session). It wouldn’t have worked if you change the ‘seasonality’ (the days around the FOMC meeting) and/or the +0.45% mark, but this would make it a different strategy. Why should one discard an otherwise favorable strategy which stood the test of time and performed extremely well in different markets because it wouldn’t have worked if you significantly vary the +0.45% mark ? For me it’s like a kid which would receive a cookie for being nice, but would discard a potential strategy of always being nice as ‘not robust enough’ because making some permutations to being ‘nice’ (e.g. being as bold as brass as being the worst permutation) would result in getting nothing at all (or even worse).

          Best,
          Frank

          • Mark says:

            I agree with Frank about bullet point 5. A lot of people say a system should work across markets in order to be robust. That’s just a matter of personal preference, though. In order to test that, you would have to find a healthy sample of systems that perform well on one and only one market, test it going forward, and then compare to another healthy sample of systems that perform well across markets. Good luck operationally defining these variables, much less stating what the variables even are (seems daunting to me, but maybe that’s just because it is 4:37 AM).

            Why is it not feasible that different markets have their own trading personalities? I can imagine a market whose trading is dominated by, say, 50 institutions. If 35 of those institutional traders use a triple EMA crossover approach then it makes sense to me that a triple EMA crossover system might have edge for this system but not for others.

            If I threw out the triple EMA crossover system just because it worked on one market but no others (that I tested) then I’d be missing the boat.

            This looks like a very insightful blog, Frank. I’m a first-time visitor and I will be back.

  7. BigBill says:

    Random question, but what version of Metastock do you use to backtest this? Also what data subscriptions do you need to do something like this. I am curious how hard/expensive it would be for me to do this sort of analysis on my own. Thanks in advance.

    • BigBill,

      I don’t use Meastock. For developing, backtesting, determination of probabilities and odds, printouts and and and I use Matlab from MathWorks. I did the programming myself. Although it took a a lot time and effort (and money as well, Matlab is not the cheapest solution), I definitely learned the most by doing it myself instead of utilizing pre-configured toolsets.

      I use several data provider (regularly commercial ones), but the majority of the data is provided by Metastock (Reuters DataLink).

      Best,
      Frank

  8. BigBill says:

    Thanks for the response. Any suggestions on where to go if one wanted to learn how to apply quant theory to their own trading? I’ve current read a few quant influenced writers besides you (e.g. Rob Hanna, Larry Connors, Sentiment Trader (via Twitter) but would like to actually learn to fish as opposed to being provided morsels of fish. Any suggestions? Again thanks in advance.

    • BigBill,

      there are a lot of good books about trading and trading strategies, e.g. (not exhaustive)
      ‘The Encyclopedia of Trading Strategies’ (J.O. Katz & D. McCormick
      ‘Trading Systems That Work’ (T. Stridsman)
      ‘Long/Short Market Dynamics’ (C.M. Corcoran)
      ‘Futures Tradin Vol. I & Vol. II’ (L. Williams)
      ‘Design, Testing, and Optimization of Trading Systems’ (R. Pardo)
      ‘Quantitative Trading’ (E. Chan)
      and and and

      Best,
      Frank

  9. sam says:

    . . . and another secret indicator, another secret system.
    My guess : in a few more days we will see a “special” offer to sell a subscription to the system for “just” $ XXX dollars.
    Sam

    • Sam,

      did you ever see any kind of advertising, link exchanges, subscription based offers or anything ‘commercial’ on my blog ? I already mentioned several times that blogging about the markets is my hobby (allowing for taking a break whenever I feel the desire to do so), not my business. Even my participation on http://www.quantwizards.com is a freelance contribution only, and I’m not compensated for in any way whatsoever.

      But would you do the same, taking a lot of effort (men-month of programming, developing strategies) and expenses (software, commercial data provider, workstation(s), …) and making all your findings public (especially without any kind of compensation), regularly on a daily basis ?

      The content of my postings is intended as a ‘food for thought’ for those interested in the markets and/or trying to develop strategies of their own.

      Best,
      Frank

  10. J says:

    Hi Frank,

    I’m in Europe over half of the year too, so great to see your blog. I am going to be honest, I find it incredible the results you have found and find them at odds in a “big picture” sense. I would gently point out for instance:

    1. The market now is far higher than it was in 1990, for it to get from 1990 to here and be higher the market would have had to gain more on up-days than it lost on down-days. Statistics also show that there are overall more up-days than down-days. Yet your strategy completely turns this on it’s head by being short all days when it is not long. You would have thought being long all days when there was no reason to be short would have been the way to go;

    2. You have a strict criteria for long days, yet no criteria for short days, but still appear to be in the market for more long days than short days. Seems strange this could happen with strict long criteria and no short criteria;

    3. I note the comment above by someone asking why you picked the 94 RSI as your criteria. I think the question posed by the other commenter can be answered factually rather than subjectively by checking the performance of all values of 90, 91, 92, 93, 94, 95, 96, 97 and 98. How do these test? Is there only a small drop off in profit as it moves further away from 94? Obviously looking at the compounding effect the difference can be significant if one trade 20-years ago was compounded…but CAGR should still only change slightly. Also what percent did that one trade return? In that sense, where it came in the data is irrelevant since an extra 10% 20-years ago will get you to the same overall total profit now as an extra 10% would last year. How does this one trade compare with the average return of that criteria when it is isolated from the rest of the strategy?

    4. With respect to comparing using RSI 94 instead of say 95 and going long a FOMC day, a day earlier or later, I think it is comparing apples to oranges. They are totally different trades and the structural reasons for them working are totally different. FOMC has many institutions positioning themselves for the meeting and generally in the same way. In this circumstance it is the day that is important, too early and the institutions won’t be there to assist the trade, too late and they will have already made the move. But RSI 94 (not just because you link it with a proprietary indicator that no-one else has, so not the same as <45%, which everyone would know) is not the same and numbers around it should be tested to prove a smooth drop off in profitability either side of 94;

    5. As I mentioned, I am surprised any strategy is able to make a profit while being short by default. But 65%+ CAGR with 16% drawdown and NO leverage…well that is incredible, as is doing it with a profit factor of only 1.68 (I know it is hard to tell just from this when compounding returns as the last few trades can change this significantly, but it still seems low);

    6. I have coded many strategies that actually make a lot of money in the market but if I had come up with these results, I would check and double check the code for errors and the code "looking forward" to data not yet known at the time of the trade. Even simple things like whether the money management part of the strategy is actually investing only what you really have in cash on each trade. Without compounding, how does the strategy look with a fixed $100,000 being invested on each trade? This strategy is definitely on the right track, much of what you have chosen for the strategy does really work…just in my experience it has never worked as well as your data is showing and even more so by being short the market by default;

    Sorry for the long comment, I really admire you for putting your ideas out there for comments. I hope you will take the comments as fruit for discussion and with the spirit they are offered.

    All the best,
    J

  11. greattest says:

    first excellent work.

    how would the strategy have fared in 1987 crash? i think that is important to check. perhaps worth getting access before going live with this.

    and other problem is that since it is based on a large amount of trades, i think commissions, margin requirements for shorting, etc should also be analyzed before u go live.

    that being said, my guess is you already know all of that and i think the work you have done is excellent so far

    • greattest,

      thanks a lot for your kind words.

      I think commissions are not an issue due to the fact that the strategy only takes a position once a day, and commissions only apply when positions are switched (long to short or vice versa). On all other days is a ‘buy and hold’ approach (only for statistical purposes daily changes are reported). Commissions are currently at 0.0044% per share ($0.005). But my next stats will include commissions, fees and slippage.

      I don’t have all the breadth data necessary before 1990, but that wouldn’t matter. A -20% day (an outlier) wouldn’t make the strategy unprofitable (in fact is’t nothing else than losing -2% on 9 additional days over the course of several thousand sessions), and even if positioned correctly on the short side performance stats would look even more impressive.

      Best,
      Frank

  12. david says:

    Frank,

    would you be interested to publish you model dashboard ( signal, history, etc. ) ?
    I’ll create a website with automatic updates for this

Leave a Reply

Your email address will not be published. Required fields are marked *

* Copy This Password *

* Type Or Paste Password Here *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>