Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Simple RSI & EMA high Profitable ratio Strategy 89 replies

Time action momentum trading 90%+ accuracy 0 replies

Win/Loss Ratio and R/R Ratio Link? 11 replies

Trading with Coinflip Accuracy!! 29 replies

iran calls package acceptable - n. t. 3 replies

Attachments: Acceptable Accuracy Ratio for Profitable trading
Exit Attachments

Acceptable Accuracy Ratio for Profitable trading

  • Post #1
  • Quote
  • First Post: Edited 8:39am Dec 10, 2017 7:36am | Edited 8:39am
Hi.

I am working on a set of systems that guarantee a certain proportion of successful forecasts. The idea is that knowing the number of times the system guesses correctly the next candle prices and how many times it fails, you know perfectly your risk. Knowing your risk, you can now establish a strategy based upon total profit and loss according to your take profit and stop loss levels.

So, here my question for those who are experienced traders, since I am more experienced on mathematics is: which percentage of accuracy would you expect from a system to make it profitable?

For instance, working always on the Daily time frame, for next day close prices, my first model gets around 75% of correct forecasts. For next day open price 87% of correct predictions. I have not been working yet on high and low (total daily price range) because those require more work around. So let's focus on next day close price.

Do you think 75% of successful predictions is profitable, which is your average ratio and what is the profitability your systems based upon it and their timeframes?

Cheers.
I managed to create trading systems with swarm optimization,evolutionary algorithms with % 90 plus winning percentage of accuracy.But they still produced NEGATIVE RETURN ON TRADES.So the winning percentage is not a good criteria to build a trading model that last.My experience is trading systems that last long should produce signals at close to peaks and valleys.
Quoting patara
I managed to create trading systems with swarm optimization,evolutionary algorithms with % 90 plus winning percentage of accuracy.But they still produced NEGATIVE RETURN ON TRADES.So the winning percentage is not a good criteria to build a trading model that last.My experience is trading systems that last long should produce signals at close to peaks and valleys.
I get your point. I have also been using a heuristic approach. The accuracy I mention is not related to winning orders, but related to knowing which is the direction the price will take in advance. I would feel comfortable with 85%, but with 75% I am still not sure if I can by playing with different supports and resistances to set entry and exit levels if it would be profitable. So the accuracy means that if I take 100 trading days, 75 of them I know which is the actual direction of prices and 25 I will miss it.

So my system (forecasting system, not trading system) is not sending any buy/sell signal, it's just giving an average guarantee that prices are moving either towards the next resistance or support. I still need to explore how is its accuracy within the two next days (or time units because it is not dependent on time frames). Some analysis also showed that the predictions succeed when volatility is high and tend to fail when volatility is low, because after examining over 3 years of daily data, the volatility found within the good predictions tended to be 3 times higher than the ones within the days on which the prediction failed. This was for me a good sign, because also that means that the stop loss ranges within those days should be smaller than take profit levels during higher volatility days.

However, it's true, and that is the reason I have post this question here and also on another site is simply that I have not tested it on detail yet and I was concerned.

I am using AI and advanced statistics because after a few years testing technical analysis and price action I do not trust in these.
Quoting scherzi
Hi. I am working on a set of systems that guarantee a certain proportion of successful forecasts. The idea is that knowing the number of times the system guesses correctly the next candle prices and how many times it fails, you know perfectly your risk. Knowing your risk, you can now establish a strategy based upon total profit and loss according to your take profit and stop loss levels. So, here my question for those who are experienced traders, since I am more experienced on mathematics is: which percentage of accuracy would you expect from a...
And what is the use of 75% winrate if your targets are smaller than SL's?
You say you are good at math? Then why you can't figure such a basic aspect?
Little help for you:
For a 75%winrate your R:R need to be 3:1(3pips SL:1pip TP NOT INCLUDING SPREADS) only to be at BE... Plus the commissions and swaps.... Go figure...
BTW, if your magic system will produce 65%winrate with 1:1 R:R, which I doubt with your level of knowledge, you will be millionaire in no time.

Quoting scherzi
{quote}... I am using AI and advanced statistics because after a few years testing technical analysis and price action I do not trust in these.
So you can blame anything but yourself... nice
You can trust whatever you want...it will not cancel your loser attitude.
Good luck.
For me the biggest thing that has helped my trading was do not over trade. Don't be glued to the charts all day looking for something to get into right now. On the weekend i draw my lines and get my ideas for any pairs that look like they might do something. I place alerts on trend lines so if crossed i get an email. No need to keep looking all week. If the setup i was expecting doesnt happen then well, I evaluate next weekend. But I like to get into trades near major trends if they continue boucning or break out of trend. To me that seems the least risky and has most reward.
Quoting meepstone
For me the biggest thing that has helped my trading was do not over trade. Don't be glued to the charts all day looking for something to get into right now. On the weekend i draw my lines and get my ideas for any pairs that look like they might do something. I place alerts on trend lines so if crossed i get an email. No need to keep looking all week. If the setup i was expecting doesnt happen then well, I evaluate next weekend. But I like to get into trades near major trends if they continue boucning or break out of trend. To me that seems the least...
The system I have in mind to use the forecasts is pretty similar. If my odds for a positive forecast are 75%, when the forecast fails, the next time the probability of making a good prediction is above 85%, let's say. You can same way work with weekly data on the same way. once the probability reaches an acceptable level, you check the current price setup and decide wether to enter or stay. intraday is not worth it imo, but daily and Dn offer enough volatility.
  • Post #7
  • Quote
  • Edited 2:34am Dec 11, 2017 2:05am | Edited 2:34am
Quoting scherzi
{quote} The system I have in mind to use the forecasts is pretty similar. If my odds for a positive forecast are 75%, when the forecast fails, the next time the probability of making a good prediction is above 85%, let's say. You can same way work with weekly data on the same way. once the probability reaches an acceptable level, you check the current price setup and decide wether to enter or stay. intraday is not worth it imo, but daily and Dn offer enough volatility.
Why make life difficult for yourself?

Why not trade Pitchforks or trendlines: https://www.forexfactory.com/showthr...4#post10571404

At the end of the day, the objective is to make money and there are existing methods you can use to achieve great profits without reinventing the wheel. Although the approach you've stumbled on, copied or created may have potential, is it worth it in terms of the time investment? Especially if the time can be invested in learning some solid existing method.

Using high probability forecasts is unlikely to transform a trader into a super star trader because the thing that make him fail in trading will still exist even with high probability forecasts. Unless the trader is willing to calculate the probabilities of all technical setups to use as a standard against the AI predictions how will one know if the system is relatively better or worse? Just my two cents.

Look at AlexC return and his USDJPY setup. Pitchforkers already know there is an 80% chance of the price falling to the middle blue line if price can't break the blue above price and if the price breaks and stays above the blue line there is an 80% chance of price reaching the green middle line above which is R:R > 10:1. An AI probability forecast in situations where one is using a true smart money tool do not add value, in truth they are additional noise.

https://www.forexfactory.com/showthr...0#post10578350


Anyway, good luck! Just attempting to show you there is another way because the AI journey you are about to embark on might be good for the ego but bad for the bank balance.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: 2017-12-07_173323.png Size: 30 KB
Click to Enlarge Name: 1500%.png Size: 37 KB
Click to Enlarge Name: 72b61d136dd7143a3d2b15fdd9ada9b4.png Size: 44 KB
Quoting mbrown
{quote} Why make life difficult for yourself?...
When it comes the time to develop a trading strategy price setup is there as you conveniently show trhough trendlines, sup/res and price action. The reason I am adding probability is to better define in advance the risk I am assuming. I do not feel comfortable with single candlestick analysis so I prefer a probabilistic approach. This probability may be extended across one or multiple time units in the future to better understanding it.


Quoting dkrock
You won't like to hear this, but Houston,...
When math does not work, only chance works. Adding maths increases your expectations above the 50% level. To put it in easy works, if you are trying to work with numbers you need maths for it. However, in my opinion you are wrong, since qualitative features can be treated as a binary variable within any mathematical model. How do you think that Sentiment or Natural Language analysis is performed?
Quoting scherzi
{quote} When it comes the time to develop a trading strategy price setup is there as you conveniently show through trendlines, sup/res and price action. The reason I am adding probability is to better define in advance the risk I am assuming. I do not feel comfortable with single candlestick analysis so I prefer a probabilistic approach. This probability may be extended across one or multiple time units in the future to better understanding it.

I'm sure you're aware the above statement moves you into the realm of conditional probabilities which is an entirely different beast and a mountain of analysis, all just to workout if a break/test of support/resistance is genuine on a probabilistic basis. This can be determined by multiple candlestick analysis or simple lower high lower low peak and trough analysis. However, I wish you luck in your endeavour.
Quoting mbrown
{quote} I'm sure you're aware the above statement moves you into the realm of conditional probabilities which is an entirely different beast and a mountain of analysis, all just to workout if a break/test of support/resistance is genuine on a probabilistic basis. This can be determined by multiple candlestick analysis or simple lower high lower low peak and trough analysis. However, I wish you luck in your endeavour.
Yes, a classic approach using raw probability is not appropiate as it is not either using a linear approach such as logistic regression or bayessian inference. The model I have already developed it and as long I have time for it I will improve it. But the 75% on daily close price movement I have it already. That's not what is concerning me. I know the market cannot be explained by using a linear approach.

I think that with your answers, as long as actual traders have responded regarding percentage of wins and their profits as per their own real experience is enough information for me. I also know the theory that is on the books, that's why I asked here hoping getting feedback from actual sucessful traders regarding my probabilistic concerns.

So for now, allow me to thank you all for your responses and I am open to any further question or issue.

Now I only have open price data available since the backtesting results on close prices I have not stored it. I may upload a sample of what I am saying within the next weeks.

Regards.
Quoting scherzi
{quote} Yes, a classic approach using raw probability is not appropiate as it is not either using a linear approach such as logistic regression or bayessian inference. The model I have already developed it and as long I have time for it I will improve it. But the 75% on daily close price movement I have it already. That's not what is concerning me. I know the market cannot be explained by using a linear approach. I think that with your answers, as long as actual traders have responded regarding percentage of wins and their profits as per their own...


Not entirely sure what you meant by the above. I didn't mention it because I was being polite but the initial question you asked
Quoting scherzi
which percentage of accuracy would you expect from a system to make it profitable?
has 1000s of different answers after factoring in trade duration, profit factor, win rate etc etc; but if you're a really a mathematician (or even have UG math and computer skills) it should be a trivial task to construct a data-table in excel in 10 minutes to get the answers you seek. The excel figures will certainly be more empirically sound than seeking answers from strangers in the hope experienced traders will reply.

Writing as an experienced trader who has already walked down the path you are now starting (AI price probabilistic forecasts and the like) I am happy with a win rate of 50%, a profit factor of 2 and Return to drawdown ratio of 3:1. However, I trade 100s of different systems with wide variations because the focus is on portfolio performance rather than one system.

Look forward to reviewing the sample.
Quoting mbrown
{quote} Not entirely sure what ...
Yes, there are two main factors why I am asking it on a public forum that are not extrinsic, so maybe there the confusion:

1. To get feedback that I cannot fin on a white piece of paper because it can only come from experience.
2. To analyze how potential users of that information react to the information to analyze whether is better to apply it directly on the market or to sell that information and focusing on a distribution channel.

So using a public forum I get much more information than I would by just testing it on a Python notebook.

However, I insist, the question I ask is not related to wining and losing ratios, but to knowing in advance the direction that the price is going to take. I think that till now, no one has answered to my original question actually, even if I am getting a lot of valuable information from all your answers.

Regards.
Quoting scherzi
{quote} Yes, there are two main factors why I am asking it on a public forum that are not extrinsic, so maybe there the confusion: 1. To get feedback that I cannot fin on a white piece of paper because it can only come from experience. 2. To analyze how potential users of that information react to the information to analyze whether is better to apply it directly on the market or to sell that information and focusing on a distribution channel. So using a public forum I get much more information than I would by just testing it on a Python notebook....
From experience, you never know because you have a probability. Directional accuracy is not the same thing as predictive accuracy and this was my bind. Overtime after reviewing the directional forecast you see a pattern between price and the directional forecast, such as the directional forecast merely forecasting the direction of the previous bar (run that through Python and I'm sure you'll have a fabulous directional model) but it didn't add much value for me from a trading perspective because I could glean that directional information from looking at the direction of the previous bar. Using EURUSD daily as an example, it has many more consecutive up and up than up and down, learning that alone will give a model a high directional accuracy but it doesn't remove any of the uncertainty from trade execution, no more than betting in the direction of the previous bar. I hope that answers your question.

I'll add...it actually adds more noise to the trading decision at least that was my experience.
Quoting scherzi
{quote} I think that with your answers, as long as actual traders have responded regarding percentage of wins and their profits as per their...
No, it's not : you need to understand & model the expectancy of your system or method.

Van Tharp's "Trade your way to financial freedom" will get you started.

Trying to "know things in advance"will hold back your progress.
Thank you, I will check that book, I am unsure if I have read it in the past.
Quoting scherzi
Hi. I am working on a set of systems that guarantee a certain proportion of successful forecasts. The idea is that knowing the number of times the system guesses correctly the next candle prices and how many times it fails, you know perfectly your risk. Knowing your risk, you can now establish a strategy based upon total profit and loss according to your take profit and stop loss levels. So, here my question for those who are experienced traders, since I am more experienced on mathematics is: which percentage of accuracy would you expect from a...
when you can work out how to be profitable at any win rate, even at 5%, over the long run then you have mastered the game
You came here for the truth and let me unveil it for you
Quoting RondaRousey
{quote} when you can work out how to be profitable at any win rate, even at 5%, over the long run then you have mastered the game
What's the point for doing that
Quoting EcoTrader
{quote} What's the point for doing that
To be profitable... No? LoL
Why not treating the prediction(predicted values) as classification problem to find out the accuracy of profitability of your system ?
For example when you predict the daily candle 75% time correct ,how much the predicted daily candle move in pips?I am going to send you a pdf file by private message, check your box.
  • Post #20
  • Quote
  • Edited 4:32am Dec 12, 2017 3:59am | Edited 4:32am
Quoting patara
Why not treating the prediction(predicted values) as classification problem to find out the accuracy of profitability of your system ? For example when you predict the daily candle 75% time correct ,how much the predicted daily candle move in pips?I am going to send you a pdf file by private message, check your box.

Why not upload the pdf so we all can read it? (super secret??? )
0 traders viewing now