Many of MS users bent their head against the wall trying to code some recursive filters that are so abundant in the trading literature. I will split this post in 2 subjects: 1) The "+PREV-PREV" construct explanation If one of the MSFL experts can provide some details about how this construct works and few examples about what can be achieved with it. (one example is from the Equis Customer Zone web site "Tips from TASC Magazine" = Ehlers filters) 2) Recursive Equations An example about the CORRECT way to implement this type of recursive code: f:=a*Typ()+b*f[1]+c*f[2]+d*f[3]+e*f[4]; where f[1],f[2],f[3],f[4] are the referenced function values 1,2,3 and 4 bars ago. (It's clear that PREV=f[1] but I believe the Ref(PREV,-1) is not equivalent to f[2]). I am sure this post will provide an extremely valuable answer for the MS community, so the experts of MSFL that can find a solution for it will be highly praised. Guara

I don't think I would use the word "locked" :lol: in reality what happens by using +Prev-Prev is that you are forcing the software to start its calculations for each bar from the first bar of data available to the current bar ... The lastvalue fools the software into believing the array is constant. That's why it works and that is why it makes calculations so much slower ... I honestly do not believe it was intended by Equis, I just think they were lucky to find this ... Patrick :mrgreen:

Thank you Jose and Patrick for your answers. 1) The "+PREV-PREV" construct Here is my sequence of tests for the "+PREV-PREV" combination: z1:=LastValue(C+PREV-PREV); z2:=C; z1==z2 at any point in the chart. The 2nd test has proved also that the following 2 equation generate exactly the same output, where "x" and "y" are listed in your CCI formula: z1:=Abs(x-LastValue(y+PREV-PREV)); z2:=Abs(x-y); z1==z2 at any point in the chart. In the 3rd test is where the difference shows up: z1:=Mov(Abs(x-LastValue(y+PREV-PREV)),pds,S); z2:=Mov(Abs(x-y),pds,S); z1<>z2 for almost EVERY point in the chart; So my conclusion is that the result of the "+PREV-PREV" construct depends on the context in which it is used. I accept Jose's explanation as a valid one that comes from a very experienced MSFL user, but I must agree that the construct is a very elusive one and its behavior isn't obvious at all. To recap: z1 in my 3rd test in fact calculates the SMA of an absolute difference between Typ() and its own SMA while its own SMA last bar value is kept CONSTANT for a BACKWARD time period equal to the length of the SMA. This behavior is the result of the "+PREV-PREV" construct in the LastValue function. The difference between z1 and z2 is that z2 calculates the SMA of the absolute value of the difference for EACH SMA value at EVERY bar, instead of keeping it constant for a BACKWARD time period equal to the SMA length. Jose please confirm. 2) Recursive equations Any ideas if this is possible? Thank you Jose and Patrick for your inputs Regards Guara

Here is another example of odd behavior: ti:= 15; pr:= MP(); {test 1} coef:=Sum(Power(Ref(LastValue(pr+PREV-PREV)-pr,-1),2),ti); test1 := Sum(coef*pr,ti)/Sum(coef,ti); coef:=Sum(Power(Ref(LastValue(pr+PREV-PREV)-pr,-1),2),ti); { test2 } coefA:=(Power(Ref(LastValue(pr+PREV-PREV)-pr,-1),2)); coef := Sum(coefA,ti); test2 := Sum(coef*pr,ti)/Sum(coef,ti); test1;test2; Once again, the context is different, and so the result is different. There is a divide by zero error, and it plots equal to zero, except for the last bar. Brad Ulrich www.thedml.com

I finaly got it by myself! A good example is posted in the "Formula Collection" under the Forum Index or in "Formulae": - "ITrend by Dr. Ehlers". In short this is the general equation I was asking about: f:= A0*Pr + A1*Ref(Pr,-1) + A2*Ref(Pr,-2) + A3*Ref(Pr,-3) + A4*Ref(Pr,-4) + B0*PREV + B1*Ref(PREV,-1) + B2*Ref(PREV,-2) + B3*Ref(PREV,-3); where "Pr" is the PRICE type of choice (i.e. OHLC,MP(),TYP(),WC() or a short MA). But be aware of the computing penalties, especially when you are using this type of equation in simulations! It's my guess that it can take easily days for just a few hundred tests. Make sure you have a good CPU fan when you intend to run backtesting with this animal! This is when the DLL's external functions are coming in handy. Guara

Roy I don't have your MSFL expertise but the decision to spend the additional time to find a more "efficient" solution to implement the same thing without PREV is very easy for me to make: - pay a MFSL expert like you or Jose to see if there is a solution at all OR - wait for a DLL that provides the function My perception as a "young" trader is that one must spent his valuable time in developing and testing TRADING SYSTEMS, not becoming an MSFL specialist in order to overcome the CRONIC MetaStock Formula Language limitations! If one has to learn programming or improve his own programming skills, my approach is to go for C++ or Delphi, buy the MDK developer's Kit from Equis and write your own DLL. It's one time learning curve and it will always provide the user with the best solutions. I have all the respect for you and Jose, don't read me wrong, but personally I don't believe there is any future for the current format of MSFL. I don't know how others deal with these kind of issues and MSFL limitations, but I will not spend my time banging my head to the walls in trying to find a work around by using what I perceive to be a DEAD LANGUAGE. As regarding the ITrend solution I've posted, this was done in order to provide an intermediate support for those interested to evaluate the indicator and later on - maybe - to buy it from "The DML" company (the ADSI package). Regarding my general recursive function formula, all the Ax/Bx are constant coefficients, or have a very simple formula to calculate them using the basic 4 arithmetic operators (+,-,*,/). I have no idea how a RECURSIVE FIXED DSP Filter formula can be re-written without using a self-referencing equation of the original function, or if it's possible at all. These formulas are generated in most cases by using dedicated filter software. I got my hands onto one of these "wonders" that generates FIR filter coefficients with very efficient attenuation in the cut-off band, but guess what: MetaStock is not able to handle constants with 13-15 digits after the decimal point! Not to mention few hundreds of Ref() instantiations necessary to avoid ther use of IIR filters that have to use PREV! I know there is a very capable man (Grant Robertson - Product Development Manager) that took on his shoulders a very challenging task that is to bring MetaStock to a competitive level with the other most popular platforms in the industry. I hope all his efforts and the contribution of dedicated people like you and Jose will make his task easier. Regards Guara

Jose I appreciate very much your input. I am inclined to agree with you regarding a complex trading system: I don't believe the more complex the system the better. Quite the opposite. But to disqualify the advances of science and its increasing capability to provide new models for the trading market I cannot agree with. I truly believe the more one can read, the larger his understanding grows. The legendary methods of J.W.Gann, the Fibonacci series of numbers, the revolutionary cycle analysis based methods of J.M.Hurst have proved that mathematics, physics and other scientific domains did and can provide valuable break-through models and solutions for the trading markets. I am not discounting the occult sciences as well as planetary harmonics, cosmic cycles, other universal forces etc. I do not intend to open an argumentative session about this matter. I am just saying that for those technically inclined and willing to use scientific methods in trading, the current MS Formula Language is not the right solution. On the other hand, there are many solid and classic consistent systems that are not using "high-tech" functions that do provide very wealthy returns to those that taught themselves how to use them, and most importantly how to exercise a solid trading discipline. For this last mentioned group I agree the MFSL is quite sufficient. Regards Guara