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

Actually I misinformed you, its slightly more complicated than that, its a simple average worked out from the previous 2 days mid point price, I will try to recreate it again, cant find my spreadsheet. If I remember correctly I looked at VPs explanation of the NI accuracy wording and worked it out from that.

On your other point I could well believe they coded it up and called it a neural index, they coded up a standard MA and called it 'predicted', not to mention their predicted stochastics and MACD that are not in any way predictive.
OK here it is, 250 days tested on VPs EOD data for the mini DOW, NI replicated every day except 11, overall 96% correlation to VPs NI. Spreadsheet is in open office S/W so havnt posted it but you can easily replicate it.

Worked out as follows using data from VP excell export.

1. Work out the average price for each day (daily high + daily low)/2

2. Work out the previous 2 days combined average price (average price yesterday + average price day before yest)/2

3 If todays close is greater than point 2 above then the 'homemade neural index' is 1, if it is less than point 2 above then the 'homemade NI' is 0.

So no pork bellies, corn or other markets examined and no 3k price tag, 96% the same as what VP puts out, maybe someone can explain how that can be. Perhaps I'm a genius like the guys that wrote VP although I think there is a simpler explanation.

Further work may result in a higher correlation, maybe even 100% but then why put the work in as its useless for trading anyhow.
Quote
OK here it is, 250 days tested on VPs EOD data for the mini DOW, NI replicated every day except 11, overall 96% correlation to VPs NI. Spreadsheet is in open office S/W so havnt posted it but you can easily replicate it.

Worked out as follows using data from VP excell export.

1. Work out the average price for each day (daily high + daily low)/2

2. Work out the previous 2 days combined average price (average price yesterday + average price day before yest)/2

3 If todays close is greater than point 2 above then the 'homemade neural index'...
I did this same kind of test long ago when I had/tried to use vp and got pretty much the same result. I tested it on almost all of their forex pairs for multiple years worth of eod data. At the time, it was based on the close of the bars instead of the mid-price of the bar, however it doesn't really change the results.

The reason you can come up with almost the same exact signal on a simple excel spreadsheet is that most people who buy it won't study/test it enough to see what it is actually calculating. And that's what they're counting on. They created an indicator that shows great statistical results, but is pretty much useless for trading.
Making my living one pip at a time.....
Quoting cp_trader
I did this same kind of test long ago when I had/tried to use vp and got pretty much the same result....
I think they probably change the calculations every now and again and sell it as an upgrade. Same as the MAs which change every now and again. So were you able to get 100% correlation? Personally I think if we can prove it does not do intermarket analysis as claimed then there is a very good case for legal action.
Tick, A question to make sure I understand,...

you said,

1. Work out the average price for each day (daily high + daily low)/2

2. Work out the previous 2 days combined average price (average price yesterday + average price day before yest)/2

3 If todays close is greater than point 2 above then the 'homemade neural index' is 1, if it is less than point 2 above then the 'homemade NI' is 0.

----------------

at step 3., are you saying if today's closing price is 0.2 above (or below) - 'what'-? The average you calculated at step 2.?

If that's what you mean, then what you're looking at is today's closing price relative to a 2SMA of the previous 2 days average prices, and not including today's prices in the calculation, right?
ok, just for reference, here's how the VP software defines the neural index:

--------------------------------------------------
The Predicted Neural Index (PIndex) is a proprietary indicator that predicts whether or not a 3-day simple moving average of the typical price (average of the day’s high, low and close price) will be higher or lower two days in the future than it is today.

The Predicted Neural Index compares two 3-day moving averages to one another – today’s actual 3-day moving average with a predicted 3-day moving average.

When the predicted simple 3-day moving average value of typical prices is greater than today’s actual 3-day moving average value, the Neural Index is “1.00,” indicating that the market is expected to move higher over the next two days. When the predicted simple 3-day moving average value of typical prices is less than today’s actual 3-day moving average value, the Predicted Neural Index is “0.00,” indicating the market is expected to move lower over the next two days.

The Predicted Neural Index is either correct or incorrect so its performance can be measured in terms of percent correct to produce the accuracy statistics cited for VantagePoint, which has an average predictive accuracy of about 80% across a wide range of markets.
----------------------------------------------------
Quoting 89tick
OK here it is, 250 days tested on VPs EOD data for the mini DOW, NI replicated every day except 11, overall 96% correlation to VPs NI. Spreadsheet is in open office S/W so havnt posted it but you can easily replicate it.

Worked out as follows using data from VP excell export.

1. Work out the average price for each day (daily high + daily low)/2

2. Work out the previous 2 days combined average price (average price yesterday + average price day before yest)/2

3 If todays close is greater than point 2 above then the 'homemade neural index'...
Tick, I ~really am not interested in arguing about this, but, your statement that "its useless for trading anyhow"... just isn't accurate. See the chart I posted earlier. I think most traders would say that is pretty far from 'useless'.

I and alot of others have posted some of the ways & times we have used the VP software to work for us, and there's really not much else we can tell you. Except that I will point out, I don't find it too surprising that your 'closing price relative to the previous 2 days SMA' approach had a 96% correlation with VP's 'todays 3SMA compared to predicted 3SMA 2 days from now' - as you are comparing ~averages~ of relatively small price movements, over relatively short periods of time - and so of course they are going to be similar.
Quoting Tango46
Tick, A question to make sure I understand,...

you said,

1. Work out the average price for each day (daily high + daily low)/2

2. Work out the previous 2 days combined average price (average price yesterday + average price day before yest)/2

3 If todays close is greater than point 2 above then the 'homemade neural index' is 1, if it is less than point 2 above then the 'homemade NI' is 0.

----------------

at step 3., are you saying if today's closing price is 0.2 above (or below) - 'what'-? The average you calculated at step 2.?

If...


Step 3 is to check whether todays close is higher or lower than the average price you get after step 2 (thats what I meant about point 2 above, I didnt mean 0.2). By the way I'm not sure that the same as a 2SMA.

Different versions of VP may have used a slightly different calculation but my point is that all the information required for the so called NI is contained within the high and low prices of previous days, not in any complex intermarket relationship, or at least if it is its a hell of a coincidence!

As regards whether or not that is of use for trading it's true that for periods of time, when things are trending, it will be sufficient. However I backtested a NI only strategy and it will fail over time, it could do well for a month or two but over time it will average out at a loss.

I suppose the other point is that if you were going to use it to trade then you (actually I should say we) shouldnt have paid 3k for it as it can be done in openoffice or excel for free.

If I can summon up the energy I'll write in into a tradestation strategy and run it over a year or two's data although I'm sure that someone has done that already.
Quoting 89tick
Step 3 is to check whether todays close is higher or lower than the average price you get after step 2 (thats what I meant about point 2 above, I didnt mean 0.2). By the way I'm not sure that the same as a 2SMA.

Different versions of VP may have used a slightly different calculation but my point is that all the information required for the so called NI is contained within the high and low prices of previous days, not in any complex intermarket relationship, or at least if it is its a hell of a coincidence!

As regards whether or not that is...
Tick, If you add yesterdays average price to the day-before's average price.. and divide the result by '2'... - you have indeed constructed the 2SMA. And in the approach you described, you are then comparing today's closing price, to yesterday's 2SMA value.

Ok, I posted the USDJPY chart earlier, with what, 3+months of neural index-only performance in 2009?, & that shows a win rate of over 80%.

For all of 2009 (6+ months so far) using that strategy, there have been 33 trades, 26 wins (79% win rate), and 21 of the 26 wins reached the 100-pip profit target.

I just went back & applied that exact same approach to the past year of USD/JPY data in VP8.0 - from 06/01/08 to 05/31/09 -- and note, that is thru the summer months which are known to not be so good for trading; and also right thru the great meltdown & volatility in the Fall of 2008 --

and, the results for the year were: 65 trades, 48 wins (36 with 100pips), and 17 losses

Thats a win rate of basically 74% (73.8%). I'm sorry, but that just doesn't seem to me to be something that fits the bill of "do well for a month or two but over time it will average out at a loss"....

And, like Forest Gump, "thats all I have to say about that".
Interesting, what stoploss and target did you use and did you scale out? I didn't get a chance today but if you feel this could be profitable I will spend some time coding it in tradestation. That gives us an automated backtesting ability. Even so I still say that the NI has nothing to do with intermarket analysis

ps I assume you only trade on the days the NI changes hence the small number of trades.
Here's the definition of the Neural Index from the Vantage Point website. It's in the little small gray print at the bottom:
"* VantagePoint's accuracy statistics were computed on out-of-sample price data utilizing neural networks trained on both single market and intermarket data and relate to the Neural Index which indicates whether the average of tomorrow's typical price and the typical price of the day after tomorrow (both unknowns at this time) are expected to be higher or lower than the average of yesterday's typical price and the typical price of the day before yesterday...."

So you have 5 days total that are involved with this indicator,

Day1: The day before yesterday's typical price
Day2: Yesterday's typical price
Day3: Today's typical price, the day you get the signal of 1 or 0
Day4: Tomorrow's typical price "unknown"
Day5: The day after tomorrow's typical price "unknown"

This means that when you get Today's signal of 1 or 0, that the average of the next 2 days (Day4 and Day5) will be greater or less than the average of the previous 2 days (Day1 and Day2).

If you put the EOD data into a spreadsheet, you can see that if today's typical price is greater than the average of the previous 2 days, then the signal is almost always a 1, and if it is less than the average of the previous 2 days, then the signal is almost always a 0.

The reason this isn't necessarily helpful is that the average of the next 2 days is already more likely to be greater or less than the the average of the previous 2 days whether it goes up or down simply because the current price is already greater or less than the average of the previous 2 days

So basically if you get a signal of 1, the prices of the the next 2 days could both go down, and as long as their average price isn't less than the average price of the previous 2 days, the PNI will still have been correct.

Hope this made sense. I can try to clarify more if you'd like.
Making my living one pip at a time.....
Precisely, although the real deception is claiming to use intermarket analysis when it clearly doesn't.
Quoting 89tick
Interesting, what stoploss and target did you use and did you scale out? I didn't get a chance today but if you feel this could be profitable I will spend some time coding it in tradestation. That gives us an automated backtesting ability. Even so I still say that the NI has nothing to do with intermarket analysis

ps I assume you only trade on the days the NI changes hence the small number of trades.

Cool, Yep, only placed trades when the NI changed direction. SL placed 20pips above/below the analysis day's high/low, depending on if going short or long. No scaling in or out, and the exit criteria was to exit at 100pips profit, or first daily profitable close, whichever comes first.

What version of VP are you working with?
Quoting cp_trader
Here's the definition of the Neural Index from the Vantage Point website....
Sure it makes sense. I ~wish the SW would give us a prediction based on closing price, rather than 'typical' price -- that would make it clearer. I think they used to? not sure. And not sure why they changed.

(BTW, the definition of neural index i posted earlier was copied/pasted right out of the help section in V8 of the SW)
Quoting 89tick
Precisely, although the real deception is claiming to use intermarket analysis when it clearly doesn't.

I don't understand how you can make this statement Tick -- just because you found a 'price vs SMA approach' that works 'kinda' or even 'alot' like the NI - it doesn't logically prove or substantiate that 'no' intermarket calculations are going on, does it?

I've built & used neural nets before- and I've seen them interpolate on high dimensional data - meaning data with alot of independent inputs, looking for a single output -- and when trained properly, the nets are able to 'learn' the data & the shape of the surfaces between data points, in this high dimensional space, even ones way beyond what we can visualize in 3D or even say 4D -- and provide estimates that are pretty darn good. (if VP is using 25 other markets to predict MAs etc in 1 market - then it is working in a 26D space!)

So, I know & believe in 'the technology' - and I have to say, I just haven't seen any evidence that VP ~doesn't~ have 'intermarket analysis' going on. With that being said, I do agree with you on how this whole discussion started - that we don't understand why in the software we don't see a 'predicted' EMA actually look different than an 'actual' EMA in the real world. I have figured out how to use the SW without really using the MA crosses anyway, but this is still a 'puzzlement'.
Quoting Tango46
I don't understand how you can make this statement Tick -- just because you found a 'price vs SMA approach' that works 'kinda' or even 'alot' like the NI - it doesn't logically prove or substantiate that 'no' intermarket calculations are going on, does it?

I've built & used neural nets before- and I've seen them interpolate on high dimensional data - meaning data with alot of independent inputs, looking for a single output -- and when trained properly, the nets are able to 'learn' the data & the shape of the surfaces between data points, in this...
Well it makes no sense that you can replicate the NI so accurately without doing any intermarket analysis, especially given the claims as to the complexity of how its calculated. Add to that the fact that they just use normal MAs and indicators like MACD and stick the word predicted in front of them and I think I'm entitled to be sceptical. Any time I asked VP about how to use the software I got some vague response about how you might want to look at this or that combination of indicators.

If you can use it then good luck, but you have to admit its a very simple set of indicators that are massively overhyped.

By the way remind me what stops you use on the NI only method, I have started to write something up in Tradestation.
Sure, Post #1233 above has the stops & tp logic
OK thanks, somehow I missed your original post on the subject.
Having some difficulty in fully automating this in Tradestation but the following looks promising. By the way Im calling the 2SMA calculation the Non Neural Index (NNI) seeing as how it doesnt analyse any other markets.

Long Trades

1.Slow stochastic is oversold on the analysis day
2. There is a bullish PIN bar or Bullish Outside Bar (as described on james16 thread on FF) on the analysis day
3. The NNI has changed from 0 the previous day to 1 on the analysis day

Buy at close on the analysis day, set stoploss at low of analysis day bar, set take profit at 1:1 ratio of stoploss (if stoploss is 120 pips then TP at 120 pips). I did this to save working out stop levels for each pair and the 1 to 1 ratio is probably the minimum we should be aiming for in any case.

Also if stochastic is trending up (like if its at 50 or something) and conditions 2 and 3 are met then you can also take a long. Shorts are opposite of long trades.

The trades do not occur frequently but win loss ratio is way better than 50/50. Problem is that there are very few trades so I have included all available forex pairs and Im looking for other 23 1/2 hr a day markets to try out, preferably ones that dont close after my bedtime!

Based on this I shorted GPBUSD yesterday. I will post a screenshot if anyone is interested.
Tick,

How about showing us a chart with trades on USDJPY based only on your NNI alone, like I did with VP's predicted neural index? The time period on that chart was 12/31/08 to 03/17/09

If you can just generate the entry signals from the NNI and send me a list of dates, I'll could overlay them on that USDJPY chart & post it.

GoodOnYa shorting GBPUSD yesterday. VP's neural index called a short the day before yesterday (evening of 30June) and has stuck with that, and also called a long on USDJPY the day before that (evening of 29June).

Cheers
0 traders viewing now