Thanks Eugene. Perhaps it is a setting in my Eclipse that is the cause of the problem. Even though I ran a backtest on a full day's data (10:00 AM to 15:30 PM), I only got the last 800 timestamps printed out to the console or about 15 minutes of data (in 1 sec. snapshots). Were you able to output the full day's worth of price snapshots?
Sorry about FAST_ERROR and PROCESS_NOISE confusion. Indeed my parameters are PROCESS_NOISE and MEASUREMENT_NOISE. ________________________________ From: nonlinear5 <[email protected]> To: [email protected] Sent: Thu, December 16, 2010 8:41:07 AM Subject: Re: [JBookTrader] Re: Status of Kalman filter? Thanks, Astor. I took your code, compiled, and tested it. Everything worked as expected: the indicator values were affected by parameters, but the price was not affected. So, I was not able to reproduce your problem. I'd note that you are referring to FAST_ERROR, but there is no such thing in either KalmanTestN.java or TensionKalman.java, which are the two files that you posted. Instead, your parameters are PROCESS_NOISE and MEASUREMENT_NOISE, and these are the ones which are passed to your indicator used in your strategy. Here is an excerpt from my test results with your code: with these parameters: addParam(PROCESS_NOISE, 10, 20, 1, 25); addParam(MEASUREMENT_NOISE, 200, 230, 1, 50000); I get this output: 1250105908000,1009.625,1009.8842495795236,1.1056038636906638 1250105909000,1009.875,1009.8840450521064,1.1056038636906638 and with these parameters: addParam(PROCESS_NOISE, 10, 20, 1, 16); addParam(MEASUREMENT_NOISE, 200, 230, 1, 10000); I get this output: 1250105908000,1009.625,1009.7869935893319,0.3920799920016001 1250105909000,1009.875,1009.790444144611,0.3920799920016001 As can be seen, the timestamps and the prices for these two samples match, while the indicator values are different, which is what one would expect. On Thursday, December 16, 2010 1:30:34 AM UTC-5, Alexana wrote: > I have uploaded KalmanTestN and TensionKalman. > > > > ________________________________ From: Eugene Kononov <[email protected]> >To: [email protected] >Sent: Wed, December 15, 2010 11:28:14 PM >Subject: Re: [JBookTrader] Re: Status of Kalman filter? > >Would you mind posting your KalmanTest.java and TensionKalman.java to the >group >so that we know that we are testing the same thing? Thanks. > > >On Wed, Dec 15, 2010 at 11:59 PM, Astor <[email protected]> wrote: > > >>I agree. That is why I am puzzled. >> >>I wanted to get a better feel for the way Kalman filter class was working. >> >>First, I took your KalmanTest.java and TensionKalman.java classes and, to >>make >>things simple, removed one of the kalman filters. So, only kalmanFastPrice is >>left. >> >>Then, I modified thegetPostState( ) method in TensionKalman, so it >>would return >>a 2-element array double[] kalmanResult, where kalmanResult[0] held the >>smoothed >>price and kalmanResult[1] held the error covariance for this smoothed price. >> >>Then, in calculate( ) method in TensionKalman, I added the following >>line: out.println(price + "," + kalmanResult[0]); >> >> >>I ran a backtest on the strategy, just to get the values of the price and >>smoothed price. Per your advice, I copied and pasted into Excel the values of >>price and kalmanResult[0] (i.e. smoothed price). >> >> >>When I ran the backtest again, with different FAST_ERROR value, I noticed >>that >>the price values were lagged relative to the first run. >> >>I tried with several more FAST_ERROR values and each time got a different >>shift >>in the price time series. >> >>If you repeat those steps above, you should be able to reproduce the problem. >>I >>ran the backtest on ES test file on a single date: Aug. 12, 2009. >> >> ________________________________ From: nonlinear5 <[email protected]> >>To: [email protected] >>Sent: Wed, December 15, 2010 10:10:46 PM >>Subject: Re: [JBookTrader] Re: Status of Kalman filter? >> >> >>Please explain how to reproduce the problem. The price should not be affected >>by >>indicator parameters. >> >>On Wednesday, December 15, 2010 10:10:18 PM UTC-5, Alexana wrote: >> >>>I see a very puzzling behavior in JBT, which may be important. >>> >>>I have been looking at Kalman filter as a replacement for EMA. The parameter >>>Measurement Noise serves the same function for Kalman filter as Period for >>>EMA. >>>So, I am optimizing the Measurement Noise to control the level of smoothing >>>that >>>the filter provides. >>> >>>As per Eugene's suggestion below, I have added the following line to the >>>calculate( ) method: >>> >>>System.out.println(price + ","+ kalmanResult[0]); >>> >>>I then copied the output and pasted it into Excel. The puzzling thing is >>>that >>>the values of price changed when the parameter value is changed from the >>>backtest GUI in JBT. Of course I would expect that the value of >>>kalmanResult[0] >>>would change in response to changing parameter, but the values of price >>>should >>>remain constant. If the values of price are affected by changes in the value >>>of >>>the parameter, then the backtester may not be working properly. >>> >>>I used ES test file in my backtest and ran the backtest for one day only, >>>Aug. >>>12, 2009. >>> >>> ________________________________ From: nonlinear5 <[email protected]> >>>To: JBookTrader <[email protected]> >>>Sent: Mon, December 6, 2010 4:36:13 PM >>>Subject: [JBookTrader] Re: Status of Kalman filter? >>> >>>> Eugene, I have added the process noise to your code and am trying to >>replicate >>>> the graphs for Kalman and EMA that you have posted below. Is there a >>>> feature >>>in >>>> JBT to create those graphs or did you have to export the data to external >>>> package? >>>> >>> >>>No, there is no direct support for that in JBT. What I did was simply >>>print a comma-separated list of indicator values from the >>>onBookSnapshot method, and then copy and save the output in a file, >>>which can then be imported to Excel. Example: >>> double priceKalmanFast = priceKalmanFastInd.getValue(); >>> double priceKalmanSlow = priceKalmanSlowInd.getValue(); >>> double priceEMAFast = priceEMAFastInd.getValue(); >>> double priceEMASlow = priceEMASlowInd.getValue(); >>> double price = getMarketBook().getSnapshot(). getPrice(); >>> System.out.println(price + "," + priceEMAFast + "," + priceEMASlow >>>+ "," + priceKalmanFast + "," + priceKalmanSlow); >>> >>>-- >>>You received this message because you are subscribed to the Google Groups >>>"JBookTrader" group. >>>To post to this group, send email to [email protected]. >>>To unsubscribe from this group, send email to jbooktrader+unsub...@ >>>googlegroups.com. >>>For more options, visit this group at >>>http://groups.google.com/group/jbooktrader?hl=en. >>> >>> >>>-- >You received this message because you are subscribed to the Google Groups >"JBookTrader" group. >To post to this group, send email to [email protected]. >To unsubscribe from this group, send email to jbooktrader...@googlegroups. com. >For more options, visit this group at http://groups.google.com/ >group/jbooktrader?hl=en. > >> >>-- >>You received this message because you are subscribed to the Google Groups >>"JBookTrader" group. >>To post to this group, send email to [email protected]. >>To unsubscribe from this group, send email to jbooktrader...@googlegroups. com. >>For more options, visit this group at http://groups.google.com/ >>group/jbooktrader?hl=en. >> >-- >You received this message because you are subscribed to the Google Groups >"JBookTrader" group. >To post to this group, send email to [email protected]. >To unsubscribe from this group, send email to jbooktrader...@googlegroups. com. >For more options, visit this group at http://groups.google.com/ >group/jbooktrader?hl=en. > >-- You received this message because you are subscribed to the Google Groups "JBookTrader" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/jbooktrader?hl=en. -- You received this message because you are subscribed to the Google Groups "JBookTrader" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/jbooktrader?hl=en.
