• The Oversold Signals of the 20DMF

    In the weekly comment that I wrote earlier in the day, I showed a back-test study on buying MF Oversold signals calculated on the S&P500.

    http://www.effectivevolume.com/conte...-August-4-2014

    I took this opportunity to have a new look at the 20DMF OB/OS signals.

    The creation of the 20DMF

    I created the first version of the 20DMF indicator around April 2009. At that time, I already had developed an EV based 20D MF pattern for a set of stocks. The basic idea of a MF trend on a set of stocks is to short when the MF crosses below the 0 level and to buy when the MF crosses above an average, when this average line is also below 0.

    At that time, I had developed a MF pattern for about 70 sectors and it was natural to start building an indicator that would be based on an average signal of these 70 sectors.

    This is how the top panel of the 20DMF was created.

    I remember that at the time, I tested buying when the 20DMF crossed below -1% or when the 20DMF crossed above -0.75%, but neither of those ideas worked correctly. You can see below that they generated negative returns.



    I therefore tested the actual OB/OS idea, which generates signals very soon after the OS level has been reached. The idea is to combine MF/price trends on all the sectors and normalize these trends measures between -100 and +100. The OS level was fixed at -70.

    The view of the 20DMF as of mid-2009 is shown below. We can see that the OB/OS signals were much more regular than the 20DMF signals, but the most important part was that the rules for a reversal allowed the OB/OS to catch reversals very fast. This worked very well in a bearish market, when we could have very fast two- to three-day reversals. This is what I thought at that time, but in hindsight, it could be that the OB/OS indicator was simply luckier than the MF Oversold.

    We can see that more than speed, the OB/OS avoided two bad MF signals, shown in blue, and added one good buy signal, shown in orange. It is probable that the OB/OS indicator was indeed much better in the market environment of 2007 and 2008.



    In terms of returns, the difference was significant, as we can see below.



    The 20DMF and QE

    Now, if we look at both the upper and the lower panel of the 20DMF over a long term period, we can clearly see that

    1. In the upper panel, the MF sometimes swiftly moved far below the -1% level
    2. In the lower panel, since 2013, both overbought and oversold levels have failed to be hit.

    This indicates a bipolar market:

    Before December 2012, when the two waves of QE ended, everybody rushed for the exits at the same time, all sectors experienced sell-offs and the MF moved below -2%.

    Since 2013, everybody has wanted to be in the market to follow the liquidity. Now that the end of QE Eternity is really close, any event might precipitate big sell-offs. This tells us that we should look at the two big selling events of 2010 and 2011 and see if the 20DMF could easily stand these.



    Analysis of the 2010-2012 period

    In the figure below, I have indicated the MF -0.75% signals, compared to the 20DMF OB/OS signals.
    We can see that these signals were almost issued at the same time, except for deep pull-backs, indicated by the Blue colored circles. In these pull-backs, money continued to move out, even when the price had a temporary bounce.

    Since the OB/OS includes a price parameter, this indicator caught bounces and issued a buy signal that was somewhat too early compared to the MF -0.75% signal.



    We can see above that the OB/OS issued a buy signal in March 2010 that was not caught by the MF -0.75%. But even including this signal, in general, during the whole period, the OB/OS did worse that the simple MF signal.



    Below is how the two signals operated during the deep pull-backs. The OB/OS looks rather negative, but in reality, some protection levels kicked in and another buy signal was issued later on. My message here is that we could be facing a similar environment with the possibility of a deep pull-back. In such a case, it is better to ignore the OB/OS early signal and only focus on a MF signal crossing above the -0.75% level.



    Analysis of the Nov 2012 - August 2014 period

    This period, named the QE eternity, was characterized by relentless FED liquidity following. Interest rates dropped, Volatility dropped, volume dropped and dips were bought. During that period, there were only two Buy Oversold 20DMF signals, the last one being in June 2013.

    To cope with this pattern, we used intermediate buy signals that triggered when the OB/OS fell below -30 without entering into the OS zone (These buy signals are shown in blue.) We also used other "trend continuation" buy signals (Not shown here.)

    We can see below that the MF -0.75% generated more signals than the OB/OS indicator which works on deeper pull-backs. These MF signals also generated on average better returns.





    Conclusions:

    From this discussion, it is obvious that we will be better off to use this MF crossing above -0.75% as a buy signal for the ongoing pull-back. If the pull-back is shallow, then a buy signal will be quickly generated. If the Pull-back is deep, then the MF will become more negative and it will take longer for a signal to be generated.

    We will need to work more on this idea, but for now, what is important is the next buy signal.



    What about the MF signal on the S&P500?

    Below is the comparison between the Buy Oversold signals generated by the MF of the 20DMF or the MF of the S&P500. You can see that the 20DMF MF is performing better. The reason for this is simple: the MF for the S&P500 is calculated using all the 500 components. They will move with the heaviest of them. This means that the S&P500 MF calculation is slow to move. It is like a large boat whose center of gravity only moves slightly when a wave hits the boat.

    The 20DMF MF is made of more than 100 small boats, whose center of gravity reacts faster to an incoming wave.
    Comments 9 Comments
    1. 77seas's Avatar
      Pascal

      I guess you are calculating the SPX 20DMF from the daily EV of all the components and averaging them. Since SPX is a float weighted index- the weighting factors should also be float weighted. Are you using this method or is it an arithmetic average? Market capitalization of SPX components alone account for nearly 80% of the USA equity market.

      Thanks
    1. Pascal's Avatar
      Quote Originally Posted by 77seas View Post
      Pascal

      I guess you are calculating the SPX 20DMF from the daily EV of all the components and averaging them. Since SPX is a float weighted index- the weighting factors should also be float weighted. Are you using this method or is it an arithmetic average? Market capitalization of SPX components alone account for nearly 80% of the USA equity market.

      Thanks
      This is correct, but each component is indeed capital weighted and the weights are adjusted daily.
      I also replace stocks when there are changes in the S&P500 list of stocks.

      This means that the index constitution of 2009 is different from the index constitution of today.


      Pascal
    1. 77seas's Avatar
      Pascal

      Thanks for the clarification.

      However to study the behavior of past price action in SPX with EV pattern, it would be important to note the 20D MF for SPX only. Just now the table data that is available in the website is for the total market of the stocks that you follow and is probably based on arithmetical average. Presently we have considerable divergence in price behavior of small cap stocks -IWM and SPY. Would it be possible for you to generate a past 20DMF data table only for SPX in addition.

      Thanks again
    1. Pascal's Avatar
      Quote Originally Posted by 77seas View Post
      Pascal

      Thanks for the clarification.

      However to study the behavior of past price action in SPX with EV pattern, it would be important to note the 20D MF for SPX only. Just now the table data that is available in the website is for the total market of the stocks that you follow and is probably based on arithmetical average. Presently we have considerable divergence in price behavior of small cap stocks -IWM and SPY. Would it be possible for you to generate a past 20DMF data table only for SPX in addition.

      Thanks again
      OK. I'll do that as soon as I get the time. Still this week though.


      Pascal
    1. 77seas's Avatar
      That would be quite helpful and may reveal additional insights. What I read from other sources, is that this recent divergence in the performance of small cap and large cap has been seen only in 2011 when the intra-year price decline for IWM was over 30 pct. Also in the 20D MF charts we see a number for MF strength. Can you please explain this or point to the relevant link? If it is a daily thrust number - it may be helpful to have this number also in the data tables.

      Many thanks
    1. Pascal's Avatar
      Quote Originally Posted by 77seas View Post
      That would be quite helpful and may reveal additional insights. What I read from other sources, is that this recent divergence in the performance of small cap and large cap has been seen only in 2011 when the intra-year price decline for IWM was over 30 pct. Also in the 20D MF charts we see a number for MF strength. Can you please explain this or point to the relevant link? If it is a daily thrust number - it may be helpful to have this number also in the data tables.

      Many thanks
      The MF Strength is simply the MF of the day compared to the absolute average MF chnage of the past 10 days.
      100% means that the MF of the day is just at the average of the past 10 days. The positive of negative signs only refers to the direction of the MF.



      Pascal
    1. 77seas's Avatar
      Pascal

      Thanks for the explanation. So is it possible to calculate the MF strength by just subtracting the true 20DMF values for two consecutive days and then dividing by the average of the current 20DMF over 20 days?

      Thanks again
    1. Pascal's Avatar
      Quote Originally Posted by 77seas View Post
      Pascal

      Thanks for the explanation. So is it possible to calculate the MF strength by just subtracting the true 20DMF values for two consecutive days and then dividing by the average of the current 20DMF over 20 days?

      Thanks again
      Not really: You take the aboslute Daily MY of the past X days, calculate its average and then compare it with the daily value calculated at the end of the current day.



      Pascal
    1. Pascal's Avatar
      Quote Originally Posted by 77seas View Post
      Pascal

      Thanks for the clarification.

      However to study the behavior of past price action in SPX with EV pattern, it would be important to note the 20D MF for SPX only. Just now the table data that is available in the website is for the total market of the stocks that you follow and is probably based on arithmetical average. Presently we have considerable divergence in price behavior of small cap stocks -IWM and SPY. Would it be possible for you to generate a past 20DMF data table only for SPX in addition.

      Thanks again

      The new table is now in the S&P500 section.
      If you start measuring the distance to a signal, we are now at 0.25% from a buy signal.
      Note that in oversold situations (S&P500 MF below -10%,) the MF moves faster than ordinary. Hence, the daily average move is 0.24%. This means that we have about a 50% chance to issue a buy signal on Monday if the MF move is positive (50% chances). Hence, about 25% chance of a buy signal.


      Pascal

      Attachment 25391