TRADING THE ODDS

Due to the fact that when bloggers/the bloggosphere/traders are discussing and talking about the ‘overbought‘ (‘oversold‘) state of the market (betting on a short-term mean-reversion tendency), they are regularly referring to and utilizing J. Welles Wilder’s popular Relative Strength Index (RSI).

But Wilder’s RSI is not the only available momentum oscillator measuring the velocity and magnitude of directional price movements (in order to check for the likelihood of a short-term / next session’s mean-reversion tendency of the markets).

Others (list not exhaustive) are

• the Commodity Channel Index (CCI)
  (developed by Donald Lambert)
• Williams %R (%R)
  (developed by Larry Williams)
• the Ultimate Oscillator (UltOsc)
  (developed by Larry Williams)
• the DV Super Smoothed Double Stochastic Oscillator (DVSSDSO)
  (developed by David Varadi at http://www.dvindicators.com/)

Due to the fact that the original DV Super Smoothed Double Stochastic Oscillator utilizes a 10-day time frame, I applied some small adaptations to the original concept, especially with respect to the time frame, the time frame of the first smoothing period and the smoothing factor as well. So stats below and the respective bottom line may or may not apply to his original concept (additionally taking into account that my implementation in Matlab is not guaranteed to be error-free), but should be regarded more or less as a (very successful) proof of concept, even applied to a short-term time frame.

For evaluation purposes in order to quantify the short-term state of the market and check for the market’s (utilizing the SPY as a proxy for the S&P 500) historical probabilities and odds for a short-term mean-reversion tendency on the then following session, the following set of parameters (time frames) were used:

• (Setup 1) RSI(2) : the 2-day Relative Strength Index
• (Setup 2) CCI(4) the 4-day Commodity Channel Index
• (Setup 3) %R(2) : the 2-day Williams %R
• (Setup 4) UltOsc(1, 2, 4) : 1-day, 2-day and 4-day Ultimate Oscillator
• (Setup 5) DVSSDSO(3, 2, 0.55) : 3-day time frame, 2-day time frame (for the computation of the average for the first smoothing), and a smoothing factor of 0.55 for the DV Super Smoothed Double Stochastic Oscillator

For the time frame from 01/01/1990 to 02/12/2010, all indicator values were ranked from highest to lowest (day-by-day, always between the 01/01/1990 and the then current session to avoid the ‘hindsight bias‘), then utilizing the respective (then current) lowest 5th (covering approximately 253 sessions) and 10th percentile (covering approximately 507 sessions) for going long on the close, and highest 90th and 95th percentile for going short on the close checking for the market’s historical probabilities and odds for a higher / lower close on the then following session (the short-term mean-reversion tendency).

Assessment criteria will be the Growth Rate per Trade (geometric growth of cumulative returns, not annualized, but on a per-trade basis, the efficiency), the Percentage of Winning Trades (the effectiveness), the Maximum Drawdown, the percentage-wise number of Top 10% Winners and  Top 10% Losers (means the percentage of occurrences among the top 10% of the best / worst performing at-any-time sessions), and the t-score (Student’s t-test) in order to check if – and to what extent – any deviation from at-any-time returns occurred by pure chance only.

Using Stockcharts.com (chart courtesy Stockcharts.com) for a graphic representation of those indicators listed above, the run of the respective curves (below the chart) looks quite similar, but as the stats will show, the devil is in the details, and they’re ‘performing‘ (means their respective accuracy of forecast) quite differently.

Table I below shows the SPY‘ historical performance (since 01/01/1990) on the then following session assumed one would’ve bought the SPY on the close of a session where the respective indicator closed in the bottom 10th percentile of the then current range of indicator values.

For going long on the close on a session where the respective indicator closed in the lowest 10th percentile of the then current range of indicator values, both the Relative Strength Index and the Ultimate Oscillator show the best performance stats with respect to the assessment criteria defined (Growth rate per Trade,  Percentage of Winning Trades, Top 10% Winners and Losers, and respective t-scores), while the DV Super Smoothed Double Stochastic Oscillator out-performs with respect to the lowest Maximum Drawdown. But they all do a good job, with a t-score always exceeding the +2.40 mark, means there is a (very) low probability that any of these positive results occurred by chance only.

Table II below shows the SPY‘ historical performance (since 01/01/1990) on the then following session assumed one would’ve sold short the SPY on the close of a session where the respective indicator closed in the top 10th percentile of the then current range of indicator values.

With respect to gauging ‘overbought‘ market conditions providing a potential favorable edge on the short side of the market, differences in quality of forecasts between all participating momentum oscillators become quite obvious. Only Williams %R and again the Ultimate Oscillator are able to achieve (noteworthy) positive results on the short side of the market, showing compelling performance stats (and t-scores) concerning all of those assessment criteria defined, while the Commodity Channel Index (4-day time frame) completely fails with respect to timing a profitably entry on the short side.

Table III below now shows the SPY‘ historical performance (since 01/01/1990) on the then following session assumed one would’ve bought the SPY on the close of a session where the respective indicator closed in the bottom 10th percentile of the then current range of indicator values, and sold short the SPY on the close of a session where the respective indicator closed in the top 10th percentile of the then current range of indicator values (combined long/short strategy).

With respect to the 10th percentile and a combined long/short strategy, Williams %R and the Ultimate Oscillator show far superior performance stats with respect to all assessment criteria defined among all of those participating momentum oscillators, and a t-score close to or even exceeding the +5 mark is an indication that the probability that the out-performance in comparison to at-any-time probabilities and odds occurred by chance only is close to zero.

For the next run I will lower the respective percentiles from 10th to 5th, looking for the more extreme ends of the distribution of indicator values.

Table IV below shows the SPY‘ historical performance (since 01/01/1990) on the then following session assumed one would’ve bought the SPY on the close of a session where the respective indicator closed in the bottom 5th percentile of the then current range of indicator values.

Now the clear winner is the DV Super Smoothed Double Stochastic Oscillator. It shows the highest Growth Rate per Trade, the lowest Maximum Drawdown, with 4.72% by far the lowest percentage of occurrences within the top 10% worst performing at-any-time sessions, the lowest maximum losing trade and by far the highest t-score of all participating momentum oscillators.

Table V below now shows the SPY‘ historical performance (since 01/01/1990) on the then following session assumed one would’ve sold short the SPY on the close of a session where the respective indicator closed in the top 5th percentile of the then current range of indicator values.

The same picture again: With respect to the 5th percentile, Williams %R and the Ultimate Oscillator show far superior performance stats with respect to all assessment criteria defined among all of those participating momentum oscillators successfully and profitably timing an entry on the short side of the market, while the Relative Strength Index (RSI) and the DV Super Smoothed Double Stochastic Oscillator show negative (!) results concerning all relevant performance stats even on the top 5th percentile.

Table VI below shows the SPY‘ historical performance (since 01/01/1990) on the then following session assumed one would’ve bought the SPY on the close of a session where the respective indicator closed in the bottom 5th percentile of the then current range of indicator values, and sold short the SPY on the close of a session where the respective indicator closed in the top 5th percentile of the then current range of indicator values (combined long/short strategy).

With respect to the 5th percentile and a combined long/short strategy, Williams %R, the Ultimate Oscillator and the DV Super Smoothed Double Stochastic Oscillator are very close, with (again) the Ultimate Oscillator leading the pack.

Bottom line: The Ultimate Oscillator may have owned the title of the best-performing all-rounder among those evaluated momentum oscillators listed above, showing superior results on the long and short side of the market likewise with a high accuracy of forecasting a short-term mean reversion tendency of the underlying index on both sides of the market, while David Varadi’s DV Super Smoothed Double Stochastic Oscillator (with deviating time frames and smoothing factor) shows by far the highest quality of forecast (and performance stats) with respect to those market conditions regularly marked as ‘heavily oversold‘, profitably timing a short-term bottom.

At least with respect to the 10th and 5th percentiles, the Commodity Channel Index (CCI) (may be due to the fact that a 2- or 3-day time frame is not really applicable) almost always significantly under-performed in any of those percentiles (and with respect to the assessment criteria defined) in comparison to the Relative Strength Index (RSI), the Ultimate Oscillator, the Williams %R and David Varadi’s DV Super Smoothed Double Stochastic Oscillator.

Surprisingly the popular Relative Strength Index (RSI) was only able to keep up with the Ultimate Oscillator with respect to going long on the bottom 10th percentile, but significantly under-performed (to say the least) both the Williams %R and the Ultimate Oscillator on the short side of the market when the respective indicator closed in the top 5th and 10th percentile of the then current range of indicator values, vainly trying to gauge a short-term market top. Relatively high 2-day Relative Strength Index (RSI) values are – at maximum – indicative of limited upside potential on the then following session, being highly indicative of a continuation of the then current uptrend after a short (and regularly mild) pullback (see my posting ‘Overbought’ w/strong Uptrend), but not very helpful (to say the least) for timing the markets on the short side.

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.

As promised this will be the third part of How To Make Millions (in %) Trading The SPYDER – Seasonalities (I) and How To Make Millions (in %) Trading The SPYDER – Holiday Effects . The third part will deal with potential edges provided on the long and/or short side of the market shortly before or (immediately) following BLS (Bureau of Labor Statistics) news releases (Employment Situation / Job Report, and Consumer Price Index), in order to evaluate if – and to what extent- these periodical events had – and probably will have in the future – a (significant) impact on the direction and magnitude of the next session’s change.

Please take into account that the dates presented below and their respective performance figures do NOT reflect the index’ performance on the the session itself, but the next session’s performance instead, means if -and to what extent- the respective date (periodical event) would’ve provided a favorable opportunity for going long or short on the close targeting a higher/lower close on the then following session.

This time the stats take into account ‘real’ (effective) dates, no assumption is made about the ‘first Friday of a month’ (which regularly matches the release of the employment data report, but approximately 3 times a year both dates won’t match). The respective dates are available on the web at BUREAU OF LABOR STATISTICS.

xx

1. EMPLOYMENT SITUATION

Table I below shows the S&P 500‘ historical performance with respect to the Bureau of Labor Statistics’ release of the national Employment Situation (Job Report), assumed one would’ve bought the S&P 500

• Setup 1: on the close of the session immediately preceding the release date,
• Setup 2: on the close of the session immediately preceding the release date in the event the S&P 500 closed lower at least -0.50%,
• Setup 3: on the close of the session immediately preceding the release date in the event the S&P 500 closed lower,
• Setup 4: on the close of the session immediately preceding the release date in the event the S&P 500 closed higher,
• Setup 5: on the close of the session immediately preceding the release date in the event the S&P 500 closed higher at least +0.50%,

, for the time period between 01/01/1958 and 01/06/2010 (the last 52 years).

It is interesting to note that the S&P 500‘ performance on the session immediately preceding the news release is highly indicative (providing a statistically significant edge on the long and short side) for the release date’s performance, or in other words positively correlated: a higher close on the session immediately preceding the news release date is regularly followed by another higher close on the release date, and vice versa. And the weaker/the stronger the S&P 500 closes on the preceding session, the worse (lower) / better (higher) the next session’s returns (probabilities and odds) are (setups 2 to 5 show a t-score exceeding the +/- 1.645 mark for statistical significance, means there is a very low probability that the out-/under-performance in comparison to the general market occured by chance only).

Table II below now shows the S&P 500‘ historical performance with respect to the Bureau of Labor Statistics’ release of the national Employment Situation (Job Report), assumed one would’ve bought the S&P 500

• Setup 1: on the close of the release date,
• Setup 2: on the close of the release date in the event the S&P 500 closed lower at least -0.50%,
• Setup 3: on the close of the release date in the event the S&P 500 closed lower,
• Setup 4: on the close of the release date in the event the S&P 500 closed higher,
• Setup 5: on the close of the release date in the event the S&P 500 closed higher at least +0.50%,

, for the time period between 01/01/1958 and 01/06/2010 (the last 52 years).

It is again remarkable that the S&P 500‘ performance on the news release’ session is highly indicative (providing an ever better statistically significant edge on the long side) for the then following session as well, or in other words positively correlated again: a higher close on the news release date is regularly followed by another higher close on the session immediately following the release date, although this time the downside doesn’t show a statistically significant edge. And again: the stronger the S&P 500 closes on the news release date, the better (higher) the next session’s returns (probabilities and odds), with even higher t-scores.

Conclusion with respect to the market model: The model will take a long/short position on close of the session immediately preceding the Bureau of Labor Statistic’s release of the national Employment Situation in the event the S&P 500 closes higher/lower (short-term trend-following), and a long position on the close of the news release’ session in the event the S&P 500 closes higher (these findings are brand new, so at the current state the market model is short with respect to next Monday’s session although probabilities and odds are histroically tilt in favor of another higher close). At a later stage position sizing will take respect to the fact that the odds are significantly tilt in favor of the upside/downside depending on the magnitude of change (the higher/lower the S&P 500 closed on the news release session, the better for the next session’s performance) on the news release session.

xx

2. COMSUMER PRICE INDEX

Table III below shows the S&P 500‘ historical performance with respect to the Bureau of Labor Statistics’ release of the Consumer Price Index, assumed one would’ve bought the S&P 500

• Setup 1: on the close of the session immediately preceding the release date,
• Setup 2: on the close of the session immediately preceding the release date in the event the S&P 500 closed lower at least -0.50%,
• Setup 3: on the close of the session immediately preceding the release date in the event the S&P 500 closed lower,
• Setup 4: on the close of the session immediately preceding the release date in the event the S&P 500 closed higher,
• Setup 5: on the close of the session immediately preceding the release date in the event the S&P 500 closed higher at least +0.50%,

, for the time period between 01/01/1958 and 01/06/2010 (the last 52 years).

It is interesting to note that the S&P 500‘ performance on the session immediately preceding the news release is highly indicative (providing a statistically significant edge especially with respect to the short side) for the release date’s performance, or in other words positively correlated: a higher close on the session immediately preceding the news release date is regularly followed by another higher close on the release date, and vice versa.

Table IV below shows the S&P 500‘ historical performance with respect to the Bureau of Labor Statistics’ release of the Consumer Price Index, assumed one would’ve bought the S&P 500

• Setup 1: on the close of the release date,
• Setup 2: on the close of the release date in the event the S&P 500 closed lower at least -0.50%,
• Setup 3: on the close of the release date in the event the S&P 500 closed lower,
• Setup 4: on the close of the release date in the event the S&P 500 closed higher,
• Setup 5: on the close of the release date in the event the S&P 500 closed higher at least +0.50%,

, for the time period between 01/01/1958 and 01/06/2010 (the last 52 years).

As is was the case with respect to the release of the national Employment Situation, it is again interesting to note that the S&P 500‘ performance on close of the news release’ session is indicative (providing an edge on the long side, although not as statistically significant as it was the fact for the employment situation) for the session immediately following the release date.

Conclusion with respect to the market model: The model will take a long/short position on close of the session immediately preceding the Bureau of Labor Statistic’s release of the Consumer Price Index in the event the S&P 500 closes higher/lower (short-term trend-following), and a long position on the close of the news release’ session in the event the S&P 500 closes higher.

xx

______________________

For those interested in, the model’s signals will be posted from January 4, 2010 onward regularly shortly before the close (but even if posted shortly after the close, the SPY is still available for trading with good liquidity and narrow spreads of regularly $0.01) via Twitter Update by following a new Twitter account @Strategy_I . (making further improvements to the model – e.g. optimization of the short side and reducing the chance of overfitting – as well as introducing position sizing, leverage, abnormal market filters, implementing buy/sell stops and probably making the model ‘adaptive‘ will be an ongoing process in 2010)

But please be reminded that past performance is never an indication (leave alone a guarantee) for future performance, and for good reason the CFTA and NFA (COMPLIANCE RULE 2-29: USE OF PROMOTIONAL MATERIAL CONTAINING HYPOTHETICAL PERFORMANCE RESULTS) requires all members (which I’m not) to explicitly reference to limitations, risks, lack of liquidity, among others, and publish a respective disclaimer.

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.