|
PureBytes Links
Trading Reference Links
|
Dans un courrier daté du 13/03/99 04:48:52 Heure d4iver Pari52 Madrid,
K1JJ@xxxxxxxxxxxxx a écrit :
> Btw, I've traded commoditiies since 1978. Based on Day Trading the S&P 500
> realtime most every day since 1994, here's my thoughts:
>
> Though I hate rigid systems, maybe a temporary system will help to
quantify
> things somewhat better in preparation for the neural training. The reason I
> felt the fuzzy logic may help is because my normal decision making normally
> discounts certain indicators at times based on market action.
>
Yes.
Fuzzy logic interprets your indicators in a different way according to the
value of the others indicators and their respective fuzzy set membership.
For example, indic1 will not have the same meaning ( rule) dregarding to what
indic2, 3 do at the same time.
> Example: I usually buy and sell into panics. However, if I saw some
> particurlarly bullish QUIET sequence setting up, I would not look to see my
> panic indicator flash a buy signal. It may be at 50-50 instead.
>
> I am convinced in my personal trading that the markets are constantly
> changing hour to hour and day to day and that a rigid system will not
> consistantly beat the market over time...for that reason. In fact my best
> trades are when I look for the exact opposite to occur after some event.
>
> Broad Example: After a run up and then a mild correction followed by
another
> runup, I will look for a run up followed by a very sharp correction next
> time...anything to screw the majority.
>
> As for choosing the correct neural software and interfaces...I must admit
> the suggestions offered have confused me more. I can program EL well....but
> don't see the advantage of adding an external spreadsheet to use a real
time
> Neural....vs: just running TS in the charting mode as it is with the
> indicators plotted on the screen, with a neural hopefully adding a new
> "Summary" input into my thought processes.
>
The external interface is necessary to avoid the train outside TS / test
inside TS multiple sequences, if you want to build an AI predictor based on
any performance summary field.
> I'm leaning towards N-Train because of it's "low" $906 price (counting the
> TS interface) and it's real time capability. (some others as NeuralShell
> Ward are only EOD) I'm not sure about N-Train's neurofuzzy logic
> capabilities as Pierre kindly pointed out.
N-Train is a NN software.
The version that I used was a DOS based interface.
It had no fuzzy logic in it.
>That feature may be very
> important for me because of the flexibility needed...assuming fuzzies are
> not totally rigid like the neurals I've used in the past....they all blew
> out big time in real time, of course...curve fitted. I need to learn more
> about what they can and cannot do in gray areas that equates to flexibility
> of rules and parameters once trained. The LAST thing I want to do is to
> curve fit these indicators.
>
Yes.
> Pierre, based on what I've mentioned here, which module that you discussed
> would be best for my purposes? I am not sure of the differences between the
> three you offer for $4,000.... I was confused of the pricing from your web
> page....maybe tell me privately.
>
The price is really this one and includes all the tools to do the following
1) Build an ASCII file with your indicators ( TS template provided)
2) Go into Safir and load the ASCII file. All the rest is automated.
3)Once the correct sytem found, save it (as a FZB file) and go back toTS. Then
aplly the system template calling the FZB file ( that is the neurofuzzy system
built during step 2).
4)You may then retrain the FZB fle in TradeStation using an other template.
N_Train follows the same process, excepted that step 2 is not automated and
trains on RMSE error.
step 4) is not possible.
Sincerely,
Pierre Orphelin
|