Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Daily Scalping EA - Discussion Thread 349 replies

genisis data vantagepoint 1 reply

Forward Testing Discussion Thread 241 replies

VantagePoint Sample System 0 replies

VantagePoint Graphics thread 27 replies

Attachments: VantagePoint discussion thread
Exit Attachments

VantagePoint discussion thread

Quote
QUOTE=deecos]FT,
What is the basis for the “missing data periods” statement?
Take a look at the comparison spreadsheet with the three data providers. You'll notice that on various currencies there are data days that are provided in the output of one providers data but are missing in anothers. This is not the thing you'd normally notice from using a single provider because when looking at a historical spreadsheet you tend to think all trading days are there until or unless you notice that there are trading days that one provider had data for and another did not.

We are not talking about "alleged' missing data. Just take a look at the spreadsheet and you will see "actual" missing data. The outstanding issue is only "why" it is missing.

Quote
1- As I stated in an earlier posting, I should have added a disclaimer to my published research. I have done so accordingly.<o></o>
Well done. Since you don't make any money from this I'd hardly have expected a disclaimer but hey, why not?

Quote
2- I do not perceive any portion of my published analysis to be predictive. Rather, the analysis reveals historical facts that may be equated into probabilities that in turn may be exploited for the purpose of forward or back testing.<o></o>
Regardless when we use forward testing and we base our entries on predictions such as the "predicted high" then missing data from the previous day will skew the results for the next day.

Quote
Thanks for the “alert” about the “missing data periods”, I will be more wary as a result.
No problem. Incidentally none of my comments were in any way intended to cast a negative light over your posts or any data you posted. Only to notify people of the issues that may arise.

thanks for all your help,

FT<o></o>
Quoting deecos
BeachBum,
I have compiled the results of my initial set of criteria for all 13 pairs VP offers. Presumably your favorites will be included.

I am having difficulty posting all the results (timeouts) so I have included 3 result sets that can be viewed in this posting. If you want to see the remaining results for all criteria please open the attached html file (zipped of course).
Hi Deecos,

Thanks for posting your findings. I went back the data I had compiled a few weeks back and noticed I was not getting anything close to your results. While my historical dataset is smaller than yours (it spans Sep9/05 to Jan12/06), some sample results were as follows:

1) EURUSD
- # of days NI=0 preceeded by NI=1 on the previous day: 20
- # of days low reached or exceeded VP PLow: 9 (or 45% as compared to 92% for you)

2) USDJPY
- # of days NI=0 preceeded by NI=1 on the previous day: 13
- # of days low reached or exceeded VP PLow: 5 (or 38% as compared to 87% for you)

Could a smaller dataset alone account for this kind of discrepancy or did my spreadsheet get messed up somewhere along the line? Or was there maybe a string of correct predictions in a particular kind of trending market at some point?

Rather than attach my spreadsheet that might be hard to decrypt, let me post the data from the spreadsheet for EURUSD to compare notes to see what could account for these different results (all dates are NI=0 preceeded by NI=1 showing the predicted low along with the actual low the following day):

Sep 19: PLOW: 1.2072 Actual low: 1.2109 (missed)
Sep 22: PLOW: 1.2103 Actual low: 1.2040 (ok)
Sep 27: PLOW: 1.1942 Actual low: 1.1993 (missed)
Sep 30: PLOW: 1.1983 Actual low: 1.1903 (ok)
Oct 10: PLOW: 1.2022 Actual low: 1.1976 (ok)
Oct 13: PLOW: 1.1961 Actual low: 1.1982 (missed)
Oct 17: PLOW: 1.1983 Actual low: 1.1919 (ok)
Oct 21: PLOW: 1.1902 Actual low: 1.1923 (missed)
Oct 28: PLOW: 1.2006 Actual low: 1.1969 (ok)
Nov 03: PLOW: 1.1876 Actual low: 1.1807 (ok)
Nov 14: PLOW: 1.1655 Actual low: 1.1643 (ok)
Nov 16: PLOW: 1.1617 Actual low: 1.1644 (missed)
Nov 21: PLOW: 1.1675 Actual low: 1.1686 (missed)
Nov 24: PLOW: 1.1735 Actual low: 1.1711 (ok)
Nov 29: PLOW: 1.1706 Actual low: 1.1757 (missed)
Dec 01: PLOW: 1.1668 Actual low: 1.1662 (ok)
Dec 07: PLOW: 1.1670 Actual low: 1.1705 (missed)
Dec 20: PLOW: 1.1780 Actual low: 1.1803 (missed)
Dec 27: PLOW: 1.1810 Actual low: 1.1814 (missed)
Jan 09: PLOW: 1.2015 Actual low: 1.2037 (missed)

Thanks for any thoughts/insights.

BlueMeasure
Quoting bluemeasure
Hi Deecos,

Thanks for posting your findings. I went back the data I had compiled a few weeks back and noticed I was not getting anything close to your results.
The first thing to check is whether or not you are using the same data provider.

FT
Quoting fijitrader
The first thing to check is whether or not you are using the same data provider.

FT
Yes, that's a good point. I actually had a lengthy exchange with Rocket a few hours ago where we compared data in various spreadsheets that have been posted and that was one conclusion we arrived at (including the one where you put the various data provider's predictions side by side). Should the data source be the only difference then this discrepancy is surprising - maybe even alarming - since it could make or break what is an obvious edge as shown by Deecos.

BlueMeasure
Quoting bluemeasure
Hi Deecos,

Thanks for posting your findings. I went back the data I had compiled a few weeks back and noticed I was not getting anything close to your results...

Could a smaller dataset alone account for this kind of discrepancy or did my spreadsheet get messed up somewhere along the line? Or was there maybe a string of correct predictions in a particular kind of trending market at some point?...
BlueMeasure,
It was simple to re-create the findings from September 19<SUP>th</SUP> forward using my historical data, which I did. The dates returned by my query were identical to the dates you posted, however each of the dates in my result set indicated the actual low had been reached or exceeded. Upon further examination (this took a little while to get my head screwed on straight by the way) it became clear to me that my result set was in error.

The logic of my query did NOT evaluate the two preceding neural indexes as indicated; rather it evaluated the preceding day and the current day neural indexes. The current day neural index is meaningless in this evaluation because it is predicting the 24 hour period which has yet to unfold.

Not only did I double check my logic, I had a 100% independent research conducted on my calculations before I posted them in this forum. Sadly the independent researcher made the identical logic error in his calculations which resulted in a false sense of security that the results were accurate.

SO…let me come clean by saying… Previous numbers posted by me that claim to evaluate the preceding two days neural index values are faulty. As soon as I get time to re-write the logic, and get it independently tested, I will re-publish the results for all to see.

It’s not pleasant to admit an error, but err I did. It reminds me of something I say often to those around me when they make an error: “If you’re not making any mistakes then you aren’t doing anything”.

Thanks for challenging the results in a healthy way, and for contributing to the collaborative environment. I owe ya.
May the pips be with you.
Hello traders,

Just to update you on my set and forget system I posted last week. I have completed research on a few more pairs with varying results.

I am guessing that because I have received no critisism from anyone about this system so far it must be quite useful. However, I have found a discrepency with the results if anyone would like to give an opinion on it.

During the reseach I have found that the pairs that have long runs of positive/negative trends come out with the highest points. Whereas if a pair's predicted short & medium term trend jumps from positive to negative alot and never settles on one trend for any length of time the points tally is low to negative.

The hypothisis I have drawn from this is that the pair may be to volitile for VP to accurately predict on a regular basis. Therefore, for example, the Euro/GBP is very volitile and only yielded very few points over the three year period so I have decided not to trade that pair.

In addition, It also seems that there is a big difference in the number of points gained on an up trend or a down trend. Every pair I have tested so far only works on either an up trend or a down trend but not both. The hypothisis I have drawn from that is that perhaps this is because during this period the pair was on an up or down long-term trend and therefore one of them was more favourable to trade. However, what does this mean for the next three years? The trend may change, the down trend which was so successful the previous three years may now yield negative points and the up trend will yield positive points. I hope this is not the case but if anyone would like to comment on my hypothisis or have another reasons to explain these two factors I would be happy to hear from you.

Incidently, so far I have 8 pairs that are good to trade which "would have"??? brought me 16541 points over the last three years. Give or take even a thousand points for whatever reason, surly it couldn't go that badly wrong though, and you still have quite a good return. I will post all my results as soon as they are complete.

Happy trading everyone and keep up the good work...

Cheers

FoggyTrader
Deecos,

I'm glad to have been of help; I was looking at a bunch of spreadsheets for a number of hours last night and couldn't come up with an explanation for the discrepencies, so it's good to know that you were able to track down the cause of the error and I look forward to your revised numbers.

foggyTrader,

A few weeks ago, I ran a similar test to what you describe having done in your spreadsheet (I've just had a quick look at it). I actually went a step further by also using the DX (Dollar Index) predictions to further filter any currency pairs crossed with the USD as predictions were more reliable that way (e.g. EURUSD and DX are negatively correlated to a factor of almost -1). I also agree with you that knowing what the current trend is definitly helps in making better use of VP's predictions.

My initial findings for e.g. the EURUSD were as follows (for the same historical data of 84 trading days as above)

1) # of days where PTSDiff & PTMDiff are both down and the PTSDiff & PTMDiff for the DX both up
29

- of these how many was the Plow hit? (i.e. actual low was lower than Plow)
10

2) # of days where PTSDiff & PTMDiff are both up and the PTSDiff & PTMDiff for the DX down:
25

- of these how many was the Phigh hit? (i.e. actual high was higher than Plow)
7

Obviously, these aren't results one can use to key in and forget every day at 6pm EST, so I decided to take the analysis from another angle. For each day, I added a column that asked whether the phigh came to within x pips of the high, same for the plow and a third column for both. I pumped the data through a neural network using VP's indicators and some actual data as input and the columns above as output (to let the computer in a sense learn to filter out bad days when VP is off) and to try to determine what the optimal # pips from plow and phigh are for set-and-forget type trading and got results as follows:

To within 5 pips:
highlow
43%50%percentage correct

To within 10 pips:
highlow
63%77%percentage correct (this is nice!)

To within 15 pips:
highlow
60%53%percentage correct

These numbers compare what the neural network thinks of VP's daily predictions as compared to how valuable the predictions actually were (ie. x pips to within low/high as described above)

To date, after having done a fair amount of testing using VP's data, I have found these to be the most encouraging results yet. Since the historical dataset I had on hand was small, further testing needs to be done and now that I have more data from other people's spreadsheets, I'll see if I can fine tune this further (fyi Sep 16/05 - Nov 31/05 historical dataset was used to train the neural network with Dec 01/05 - Jan 12/06 (30 days) as out-of-sample data)

As always, comments/thoughts appreciated.

Good trading,
BlueMeasure
bluemeasure, foggytrader, deecos............

I have been reading and watching what you guys are doing............it's interesting, don't stop. There are, I believe, valuable things to learn about VP and how it can be used successfully to trade the Forex.

Your efforts are appreciated, and if I have something to add, I will. In the meantime, I'll continue to 'watch'......
I think it is essential to know which data provider each is using. Results will be different from one to another. You guys may have privately shared what providers you are dealing with but I think it is important for each contributor to state the source of data they are running their tests with when posting their test results. This will help readers a lot by saving us the difficulty of having to memorize who is using what data and keep track if anyone changes providers or historical data sets. It will also serve to assist people in choosing a data provider.

Also note that when there is a gap between fridays close to sundays open that the gap should be added or subtracted from the ph/pl to reset mondays prediction in VP.

FT
All,
I have corrected the logic on the back testing numbers I previously posted, and have attached the results for you to examine. The results are far less compelling than I had hoped for, so this means back to the drawing board for more complex back testing. Perhaps I will include the “strength indicator” and the PTS & PTM Diff’s to help identify more compelling probabilities. (NOTE: The results have been duplicated by independent analysis) <?xml:namespace prefix = o ns = "urn:schemas-microsoft-com:office:office" /><o></o>

If you have back testing suggestions regarding the various VP predictive numbers then post your ideas, or send me an email with your ideas, and I will compile the list of suggestions. If the list has a significant number of ideas, I will consider posting the list for all to see so a vote can be made to determine the priority of which ideas will be evaluated first.<o></o>

I intended to included the MS-Access file which includes the SQL logic, but the zipped file was just too large. If you want the MS-Access file with my SQL logic just send me an email and I will forward it to you.<o></o>

All content presented in this posting is for the purpose of information only. Any reference to past performance is no assurance of future performance. Accordingly no warranties or guarantees in respect of the content are made by the author. All content is provided in good faith and believed to be accurate, however, there are no explicit or implicit warranties of accuracy made by the author. Moreover, the author recommends that before making any investment decisions, the reader should obtain multiple opinions related to the decision and verify the facts from several independent sources.<!-- / message --><!-- sig -->
Attached File
File Type: zip 4x_predictions.zip   5 KB | 307 downloads
May the pips be with you.
Quoting deecos
All,
If you have back testing suggestions regarding the various VP predictive numbers then post your ideas, or send me an email with your ideas, and I will compile the list of suggestions. If the list has a significant number of ideas, I will consider posting the list for all to see so a vote can be made to determine the priority of which ideas will be evaluated first.<o></o>
<!-- / message --><!-- sig -->
Ok here are some ideas that I think are important:

First let me provide a context. We are looking to trade the daily range in a robust way that takes advantage of the potentially lower risk entry zone where the daily excursion of price from the mean is the entry point and the exit point is some fraction of the daily anticipated movement. The objective is to end up with a EOD trade setup with no intraday monitoring and simple self managing exit strategy.

We now have results that suggest system 4 and possibly system 2 are candidates.

The problem to be solved is basically to find out if there is a way to increase either the consistency of the system or the profitability with no change in consistency. Alternatively to simply maintiain the profitability and increase consistency. Since system four only has one flat week so far we are looking at a pretty good looking system so far.

If we reduce only a few of the days where all or most trades lose then we have accomplished something rather amazing because if we only avoided one of those days a week without missing the good trades we'd have a very spectacular system.

Therefore we may want to take a look at our rather small sample period where all trades that were entered were losers. There are several things I'd like to examine about those days:

1. Is there anything consistent about the indicators on those days?
2. Is there anything consistent about the previous two or three days that would alert us to this result?
3. For example: Let's say we use a trend def of three indicators all saying the market is long. Does the market go near the low of the day first on such days, does the market start around the median and pretty much head straight up, in other words is there any pattern to behavior that tips us off to this sort of day?
4. I'd be inclined also to want to run stats on system four exactly as it has been traded but backtesting it on 1 or 5 minute bars to approximate the OCO nature of the entries we use with that system. Currently stats have been coming from spreadsheet evaluations which do not show whether the high or low was hit first. Due to our OCO rules we must backtest with that in mind.
5. I'd like to find out if there are days of the week that rarely ever produce anthing other than losers on a pair by pair basis (not aggregate). I'd also like to find out if there are currency pairs that do not ever produce a profit ,or rarely produce one, with system 4 and also do not offer any hedging advantage through system diversity.

It's late I'll have more tomorrow,

Thanks,

FT
Quoting deecos
All,
I have corrected the logic on the back testing numbers I previously posted, and have attached the results for you to examine. The results are far less compelling than I had hoped for, so this means back to the drawing board for more complex back testing. Perhaps I will include the “strength indicator” and the PTS & PTM Diff’s to help identify more compelling probabilities. (NOTE: The results have been duplicated by independent analysis) <?xml:namespace prefix = o ns = "urn:schemas-microsoft-com:office:office" /><o></o>

If you have back testing suggestions regarding the various VP predictive numbers then post your ideas, or send me an email with your ideas, and I will compile the list of suggestions. If the list has a significant number of ideas, I will consider posting the list for all to see so a vote can be made to determine the priority of which ideas will be evaluated first.<o></o>

I intended to included the MS-Access file which includes the SQL logic, but the zipped file was just too large. If you want the MS-Access file with my SQL logic just send me an email and I will forward it to you.<o></o>

<!-- / message --><!-- sig -->
deecos...........

Thanks very much for doing this work and posting the results for us. Even though the results are not as positive as you would have liked, I think the info is useful. Keep up the good work!
Quoting fijitrader
4. I'd be inclined also to want to run stats on system four exactly as it has been traded but backtesting it on 1 or 5 minute bars to approximate the OCO nature of the entries we use with that system. Currently stats have been coming from spreadsheet evaluations which do not show whether the high or low was hit first. Due to our OCO rules we must backtest with that in mind.
FT,

We have been using OCO in our forward testing, only taking one trade per day, per pair with whichever was hit first.

I agree with the rest of your thoughts and hope we can establish some patterns. Also, I haven't looked or calculated, but it seemed like USD/JPY gave me the most fits...most losing trades, but that may have just been a predujice or something.

Thx,

drjoe
Quoting drjoe4x
FT,

We have been using OCO in our forward testing, only taking one trade per day, per pair with whichever was hit first.

I agree with the rest of your thoughts and hope we can establish some patterns. Also, I haven't looked or calculated, but it seemed like USD/JPY gave me the most fits...most losing trades, but that may have just been a predujice or something.

Thx,

drjoe
I noticed that the biggest losses come from weeks such as the most recent where the chf was trending strong but all the signals were long. The low of the day was hit first and stopped out. Strong trending is the best reason not to trade. There are two reasons why my test system was as good as it was during a time when most traders were losing:

1. Trade management of exits.
2. Trend determination to judge which side of the market to play or whether to play at all.

Before we go further though I'd like to wait for BB's totals for each system for the period. I want to see if we have any profitable set and forget systems. I also noticed activity from one strategy to another that was pretty conclusively data provider dependent. I think we need to first deal with the data provider issue to have much valid future testing.

FT
Quoting fijitrader
Before we go further though I'd like to wait for BB's totals for each system for the period. I want to see if we have any profitable set and forget systems. I also noticed activity from one strategy to another that was pretty conclusively data provider dependent. I think we need to first deal with the data provider issue to have much valid future testing.

FT
This week's results for Start 2, 3 & 5 are posted, and as soon as drjoe4x posts his, we will have the data needed to make some decisions. I do believe that data providers make a difference and comparing Strat 4 this week using two different providers will help us make this determination.
As a seperate observation: I have been using "The Knife" system taught by Dial in the PF along with VP predictions. The combination works very well. Signals to trade from "The Knife" are compared with the precicted high and low for the day. For example: if the knife indicates buy, the p/l area for the day has been reached first, the trend is up and the p/h area has not been reached............then these indicators provide me with strong reason to enter a buy. And of course the opposite scenario would work as well. But if the low AND the high have been reached when the buy signal is given, I am more cautious about entering the same trade.

These are just observations and not trading recommendations.
Hi Guys

Just some questions on the software.
First of all I dont have the software, I use something else.

Questions

When both the predicted high and low are reached/passed does it show a trend change?

When both the predicted high and low are NOT reached does it indicate a trend change?

How does the Predicted High and Low compare to MA(H,2) & MA(L,2)?

Regards
Mono
Mono
Quote
QUOTE=mono]Hi Guys

Just some questions on the software.
First of all I dont have the software, I use something else.

Questions

When both the predicted high and low are reached/passed does it show a trend change?
No.

Quote
When both the predicted high and low are NOT reached does it indicate a trend change?
No.

Quote
How does the Predicted High and Low compare to MA(H,2) & MA(L,2)?
Don't know. Try downloading a history file from the VP forward testing section and try it out.

FT
http://www.forexfactory.com/forexfor...&postcount=217
I've been following the threads involving Vantage Point and have been testing it myself unfortunately without very much success. I have been doing more backtesting than forward testing though. I have come up with something however that shows what looks like something that could be promising.

I was thinking about the 70-80% accuracy claims by vantage point and what that data is really telling us. It's saying that 70-80% of the time, it is able to predict about where the moving average will be either 2 days or 4 days into the future. Well, if this is the case, then we can use that data to tell us where the close would need to be on a given day to reach that prediction. And if we know this, then we should have a pretty clear indication of which way the market is predicted to go.

So, using the 5-day Predicted MA, at the end of the day if we take the last 4 closing prices and add them up, and then using yesterdays predicted 5 day MA, we can figure out the closing price needed to reach that prediction tomorrow using the following equation:

(Sum-of-last-4-closing-prices + X)/5 = Pred 5-day MA of Yesterday

Solve for X

If X is > the last closing price, then it is predicting a buy day, and if it is lower then it is predicting a sell day.

I backtested this on the EUR/USD using the opening price as the entry price. Using no Stop loss and closing at the close price at the end of the day, I got a result of a little over 5000 pips over 4 years. Using a stop loss of 25 pips past the PHigh or PLow I got a reasult of over 5800 pips in 4 years. I'm going to continue working on this and see if I can get better results and will post if I find any. But I thought I would share it to see what you guys think. I've attatched my results. Let me know what you guys think. Thanks.
Making my living one pip at a time.....
0 traders viewing now