To predict or not to predict

[cdas31]

Hello everybody, I am happy to found a bb like this. This is the perfect place and time at my current situation.

Preface
A friend and I started this side project where we predict the next day values for open, close, high and low of all Dow Jones listed stocks using machine learning algorithms. So, my friend is doing the math and I try to use this data to trade every day one stock of the Dow Jones. Trades will be closed everyday.

The underlying data for the prediction is using daily data of the stocks full history. My decisions per day are based on all timeframes smaller than the day chart. I want to find the correct timing for the low and high of that day and either go short or long. In one sentence, I try to trade the margin during the day and the machine helps me t find a promising stock and ideas about the entry + exit points.

Currently my personal goal is
• Win rate > 60%
• Margin per win +2%

The predictions are monitored in two steps: Technically and Systematically. The first step is that our engine drives it's own accuracy predictions and the second is a monitoring by hand. The underlying assumption is that if the predictions are at least 98% accurate they are reliable and no further testing is needed. Of course, before trading a stock this history of accuracy is taken into account.

Entry, Stop Loss & Targets
The entry is performed on time when I think the low point is reached (My software is not allowing it automated and it gives me a learning while following the chart.) Definite Exit limits are set to +2% breakthrough. Losses will be realised when breaking through -2%.

Daily steps:

1. The data is available before NYSE opens.
2. Take the stock with the highest predicted margin during that day
3. Watch the stock's open value and keep monitoring the movement
   -> If the open value is different more than 1% from the predicted value, I take this in the calculation of the expected high or low. Currently I assume that the spread will be same, but just on a different level.
4. Find the timing for entry at high or low and set target and stop loss

Disclaimer
I know there are very distinct opinions about stock predictions with machines. As I said it's a side project with which we try to prove our assumptions. Further we don't try to tell the exact numbers of the future, but to have a tool to get clues about the future - if this makes sense.

I am demo trading at the moment.

 

[cdas31]

Not really trustworthy, but still a start

To be late with posting the predicted numbers after market opens is not the most trust building action. But I am a beginner and maybe the community pardon that.

The stock with the highest predicted margin is Apple. The predicted numbers are

• Open: 104.01
• Close: 103.50
• High: 104.01
• Low: 100.62

Out of this prediction there is no doubt of going short. So, my mistake has already an impact on my performance today. That's the trade:

• Going Short early after open
• Exit after 2% (The relative predicted margin is higher than 3%, so 2% seems to be safe)
• Stop Loss = 1% (By prediction the down move is quite strong

Recap at 12:58 NY time
I traded short at 102.99 at half of the trading day. This has already been -2%.
The stock opened at 105.75 which is 1.75 points higher than the predicted open.
Considering this, the low value won't go lower than 102 in my expectation.
In result, I can be happy if during the rest of the day I can realise 1%, when low hits 102.

Update After market close
Market closed at 12.44. Not a good decision to trade after missing the opportunity.

 

[cdas31]

2nd Day - Visa

After a bumpy first day the start today is better prepared.

Most promising from Dow Jones today is Visa (V). The predicted spread during the day is 2.52 points (3.38%). The values are:

• Open: 76.38
• Close: 76.11
• High: 77.27
• Low: 74.75

Strategy is quite easy.

• Watch for the high at around 77 points.
• Going short
• Target 2% at 75.46 points
• Stop Loss 1%

 

[cdas31]

After a bumpy first day the start today is better prepared.

Most promising from Dow Jones today is Visa (V). The predicted spread during the day is 2.52 points (3.38%). The values are:

• Open: 76.38
• Close: 76.11
• High: 77.27
• Low: 74.75

Strategy is quite easy.

• Watch for the high at around 77 points.
• Going short
• Target 2% at 75.46 points
• Stop Loss 1%

Recap
I weren't able to watch the stock, because of work.

Looking back to Yesterday's movement you can say that there would have been the opportunity to go for a 1.4 % margin, because of the day range 74.85 - 75.85.

The high peak was twice in the morning and right before midday. The down peak came along during afternoon. I wouldn't had closed the trade, because the margin wasn't big enough.

In summary, not a good day, trade, opportunity.

 

[cdas31]

Last Day and China rules the world

Yesterday, was very unpredictable. That's why I decided not to trade based on the predictions.

Today I give it another shot on Visa. The predictions has been very accurate during this week (avg >98% accuracy) and the margin during today (8th Jan 2016) looks promising.

The predictions are:
* Open: 73.47
* Close: 74.08
* High: 75.13
* Low: 73.00

I think that the low peak will be reached early this day. That's why I would wait for something near 73.00 +- 0,05 points and then going long and sell latest at 75.00 points.

 

[cdas31]

Exxon for the win

Today (13th Jan 2016) the most promising stock is XOM (Exxon Mobil).

The predictions has been steady accurate (>99%) since beginning of last week.

Prediction for today is:

• Open: 74.24
• Close: 73.96
• High: 75.05
• Low: 73.18

Predicted relative spread is 2.55%, which is not the highest predicted margin today, but fairly good. Further the accurate predictions in the past support the decision to trade this stock today.

Strategy
Going long early. I assume that the stock sees its low right after the market opens. Entry at 74.00
Sell at 75.00.

 

[MajorMagnuM]

Thanks for posting, this is interesting. What is the algo taking note of in past data?

 

[cdas31]

Thanks for posting, this is interesting. What is the algo taking note of in past data?

Hey @MajorMagnuM, thanks for the feedback! The engine uses the historic data and our self build prediction model is trained with this data. The underlying technology is a neuronal network. Does this answer your question?

 

[MajorMagnuM]

Hey @MajorMagnuM, thanks for the feedback! The engine uses the historic data and our self build prediction model is trained with this data. The underlying technology is a neuronal network. Does this answer your question?

Well not quite. what aspects of the historic data does it note?

 

[NVP]

a neuronal network .........?..............

can I get one of them ?....

my one hurts these days after drinking heavily the night before :cheesy:

 

[cdas31]

a neuronal network .........?..............

can I get one of them ?....

my one hurts these days after drinking heavily the night before :cheesy:

:cheesy: funny!

@MajorMagnuM, the engine uses the stock's own historic data to predict the stock itself. Per day the values open, close, high and low are used. There is no correlation between the stocks involved, yet nor other stock related data.

That's something we are working on currently to improve the prediction accuracy.

 

[userque]

I appreciate you sharing your project here.

What NN software are you using?

 

[cdas31]

I appreciate you sharing your project here.

What NN software are you using?

Actually, it's no out of the box software. My friend programmed in R and used some standard libraries and put his own knowledge on top.

 

[cdas31]

Last week I learned a lot.
After the first weeks of trying to trade, I found that my ability to find/do good trades is not good. Why should it? I never traded before.

In addition, I learned that the prediction is not making you a good trader as well. Nonetheless it gave me a frame, direction and point to start every day.

Grasping this forum for trading strategies and what to do I found this very interesting thread of @dbphoenix, which helped me a lot on finding the right perspective on trading and the market.

 

[piphoe]

the simple act of placing a trade ....bought here means i predict price will go higher. I sold here ..i predict price goes lower.

yes, trading = prediction

 

[dbphoenix]

the simple act of placing a trade ....bought here means i predict price will go higher. I sold here ..i predict price goes lower.

yes, trading = prediction

More accurately, the results of the trader's tests provide him with the probability that price will go higher if he enters according to the criteria determined by his testing. Whether or not price will actually go higher in any given trade is unknowable. Ditto exiting and shorting.

Db

 

[piphoe]

More accurately, the results of the trader's tests provide him with the probability that price will go higher if he enters according to the criteria determined by his testing. Whether or not price will actually go higher in any given trade is unknowable. Ditto exiting and shorting.

Db

yes, that's the "edge".

that edge is really not an edge at all, but a range. edge connotes precision, there is no precision just fuzzy range thats constantly changing & evolving with price.

 

[hhiusa]

Actually, it's no out of the box software. My friend programmed in R and used some standard libraries and put his own knowledge on top.

That is pretty cool. Is there any reason why you choose the programming language R? I recently paid a friend to make a front end application for me. Are you automating the historical data download and is it in real time? You said that it looks at the OHLC only. Is the programming running once and giving you predictions for the next day?

 

[dbphoenix]

yes, that's the "edge".

that edge is really not an edge at all, but a range. edge connotes precision, there is no precision just fuzzy range thats constantly changing & evolving with price.

If it makes consistent money, it's an edge.

Db

 

[piphoe]

If it makes consistent money, it's an edge.

Db

yes. i think i have a "edge", why else place a trade?

however, my edge is fuzzy and not based on precise rules such as i'm getting out to the pip!

the markets don't work that way.