Auction Market Value Theory & Analytics

I am making assumptions based on the market data yes. Only certain conditions I have hard coded.

gotchu

I don’t know what you mean. When certain conditions exist I hardcoded them as a filter to avoid trading breakouts. Is it 100% accurate? Highly doubtful. But win rates and profitability are different considerations

can you explain what you are talking about?
We were talking about your algo

Are you questioning the results for Keppler Mapper?

Was discussing my model I thought.
I have no idea what keppler mapper is, never heard of it.
My focus is purely AMVT. Expressly the theory.

I mentioned Keppler Mapper in my previous post, to which you responded "is interesting", so I thought you might've cared to read about it.

My confusion was based on you saying that you added certain conditions as filter, as I have also used that filtering method for Keppler Mapper.

Anyways.....

You still haven't explained what you didn't understand.

So I'll assume.

"I don’t know what you mean."

I think you are referring to when I said an algo to be dynamical. Or probably the hard-coded statement I responded to: x+y = trade.

As of now, I have no clue, but I'll assume you're talking about the hard coded conditions, and how you or I think that it is not dynamic.

Let me clarify, a system that is dynamic, means, that it uses whatever strategy you have, for us that is AMVT. Now, AMVT already incorporates dynamical components. For instance, Market Condition. My previous back-test results, are an empirical evidence that, dynamically updating market regimes, in simple words, bracket/not bracket, has proved to be successful in profitability, even if the win rate is <50%.

In laymen's term, having a dynamic strategy is great!

However, I also said that it shouldn't be treated as a blind system. Market Condition being bracketing, is not a true representation of balance. Nothing can or ever will be. It's only a rough estimation.

Trades are improved via incorporating flow analysis.

However, in laymen's terms, if you have hard-coded such statements in your algo like: Volume Up 20% + Volatility 40% + 5TF Bracketing = Enter BUY.
Then, it's not a dynamic system. It's a dynamic strategy, not a dynamic algo. Two are different.

HOW?

Simply put, patterns are something that will change over-time. Hence, your algo, if it needs to be dynamic, has to actively adapt to newer patterns. That's what is called optimization. In my previous response, I said, optimization is the best way to make an algo dynamic. "I personally think a better way to optimize an aglo is to simply make it dynamic. Yes, the trade set ups will be hard-coded. However, the system needs to learn from its wins and losses. Basically, just like an ML, but with a lot more learning."

Cheers.

Dear all,
Is there anybody who knows any data provider of raw data for composite market profiles of 1, 3, 5, 10, 15, and 20 days that can be downloaded at the end of the day?

dont you guys have an indicator for this? Isn’t there something posted here? Is there no universal indicator working anymore for MT4? I don’t use MT4 so I’m ignorant of that

Krw, Yes, I do have MP Composite MT4 that we use in this forum. I just wanted to automate directly into a database instead of the slow MT4, which most of the time freezes. Where do you get your data if you are not using MT4? Any idea? Do you or anybody mind sharing any clue? Thank you

Hello JP,
Mine is custom built in excel. All I need is OHLCV data, and it does the calculations for me. After that, it's just a matter of making macros to automate organization.

I can't provide you my own code, or excel file. However, I do know people who share this data with others. But they are paid services. So I refrain from that.

If you want to, I can hook you up with them, otherwise, I think it's best that you build it on your own like I have.

Cheers

All of my data is futures data from CQG through the MT5 platform. Had to pay for the privelidge to put it on my chart. But I do try and take screenshots every day of different contracts. It has helped me learn

AMVT, Krw,
Thank you for responding. I'm making an effort to stay away from dependency on third parties. The route AMVT has traveled is so wonderful, and I am surprised you developed all the reference points just from OHLCV data. I would like to know about the service providers you mentioned. But in the end, if I get some guidance on how to do it, I would like to explore the paths of creating my own. Thank you once again. JP

Hello JP,

Appreciate your kind words.

The third party operates privately. Meaning they only deal through people they know personally. So, it's not like a fancy website or a program. You tell them how many markets data you want, and they provide it daily for $35 per month. You can get any market you want, crypto, commodities, stocks, forex etc. They'll also provide historical data if you need it. Reports (Visualizations, Charts, etc), and Analysis are also included in the package.

I bought their subscription through a common friend, and didn't had any issues with it. I left it because I didn't want to be dependent on their data, and built my own. So, if you're willing, you can message me privately and I'll provide you their email.

However, I strongly encourage to build it on your own if you can. Admittedly, it's not easy, and there aren't any universal guidelines or rules you can follow to implement this on your own in excel. Though, if you know coding, you can adapt the methods mentioned in the Cisco Articles.

The main challenge, when building this on your own, is to verify if the data you're collecting or calculating is even valid or not.

However, once you get past it, you have something very few people have in possession.

Of course, if your daily expenditure depends on trading, then you have the option for third party services. But it's not recommended for long term.

You can pay them for a while, and once you get your feet up the ground, then pay a developer to make it for you. Or the best option, is to learn it on your own.

Happy to help.

Today's EURUSD, buyers were in control.

Model Update

I've been really busy, building something that can last me a lifetime.

At-least, that's the hope.

Past few weeks was met with failure after failure. Everything, from personal life to my projects was going bad.

After trying again and again, I feel like I have something that I am able to proudly show to the AMVT members.

It's not perfect by any means, but it shows potential, and potential shouldn't be underrated.

I won't go into the depths of the technicalities of the model, nor do I have plans to make it a commercial product. That would be literal suicide.

Though, I will share the results, only for the sake of JP, who solely was interested to see. If other people are too, they're much invited to ask, explore ideas and contribute.


Briefly, I made-up something of my own. An algorithm, inspired from others. In other words, it's not a hybrid, rather, an adaptation of one to another. That's the realistic and honest way to describe what I have built.






The model was trained on 2 years of data-set. However, 30% of it was used for testing. In other words, December 2024 was used as testing month. I am working with the same data I have provided to the public.

Although, a much rigorous testing approach and large data-set is valid. And will further concrete the results.

But the proof of concept is there. It's really possible.

The model showed a genuine edge. It's proved that, via >50% win rate and high avg gain.

It's not a coin flip, lucky model. It showed authentic results.

My plans are not to make this an automated system. Rather, use it as an indicator and a helpful decision-making layer above AMVT.

Still, there's a lot of fine-tuning to do with the model. Not only that, but a larger data-set is the next required and necessary step to further criticize the model.

However, it should great results, in my opinion. It's something I created from literal scratch. No black box ML's. Pure Dynamical System. A true picture of the market.

I'll get back as soon as I start developing it more. Refining and Updating along the way.

Then, it'll be time to deploy and use it. Which is the most nervous and simultaneously and exciting part. But, as of now, it's very very promising.

I'll create a trade explorer, using this system. So the results will be as authentic as possible.

My main intention has always been to motivate research. And I still stand by that.

Hopefully these little updates and, in the near future, the results will motivate others.

See yall soon

Thank you for sharing. Looking forward to seeing the end result and the process.

Yoo guys, I have been out for couple of weeks. I have read all new threads and glad to hear you are still active. AMVT great job keep your head up, also JP for keeping up and struggling to the top. Also KRW for backtesting, really glad bcs one day evyerthing will come to its own. Believe in yourself. Besides from that, I am interested in risk management. What is your average RR or you trail TP/SL. Do you maybe hedge to protect positions? Any thoughts would be appreacited. Have fun trading

Rokica,
To be honest, I don't use Market Profile for entry, exit, and stop loss for now. But for educational purposes, let me give you a brief bit. If the market is in a rotation or bracketing, for buying, the price has to come in between the low and the LL. Once price breaks up the LL, that would be the trigger to buy, making stop loss at the low. For selling, the price has to come in between the high and the UL. Once price breaks the UL down, that would be the trigger to sell, making the stop loss the high. Taking profit would be the POC or else, if the market allows, the next swing. If the market is trending and breaks through up or down, that would be another story. As for me, I use MP to show me where the market had been, where is it now, and what is it attempting to do purposes only, at least for now. Take care. JP

Algorithm Tuning & Optimization

I was recently introduced to Timothy Masters. He's a computer scientist. Specifically, focused on trading system development.

This niche, makes him a great resource for referencing Algorithmic related tasks. Especially, when it comes to better predictions and tuning system.

I downloaded some of his books and came across a very interesting concept that I would like to share with others.

Recently, as you all know, I came up with an algorithm myself. I have never built an algo before, nor do I have plans to make an automated system. I am doing it just for the fun of it.

I was struggling with the following question: "How do I optimize the parameters and the overall architecture of the model, without overfitting?"

This is a very relatable question to anyone who has ever created an algorithm before. And it makes sense to question this.

For instance, let's say you develop a winning trading strategy. You built a model around it. You split the data-set into training and testing data-sets. You trained the model and validated it on the testing data. The model gave you >50% win rate, good profit factor, sharpe ratio etc. Overall, you think, you have developed a powerful system, as the results say so. And you may be right in your own domain.

However, the model needs optimization. And it's a necessary part. Fine-tuning a model is what you do, once you get satisfactory results.

The big challenge is that, let's suppose you optimize the parameters, change the source data features etc. You get some results that are worse, an some that are even better. You decide to keep the one's that are better. This has an extreme and a high risk of overfitting. Mainly because, you ran through the different optimizations and are validating the result on the same exact test data-set. Basically, now, your test set is what makes the model overfit, not the training data-set necessarily.

This is very tricky for new comers to find and get past it. It's what makes an algorithm look suspiciously good.

In my example, if I tested the model on a small data-set, which was 1 month of december 2024. And it gave good results. This is not because the model is trained well. It can also be a result of that month containing the necessary patterns. It can be, in other words, by chance.

Can I confidently say, if I were to test it again on november for example, would I get the same or better results? If anyone has understood what I said so far, it's highly unlikely that I can get even close to what I got in december.

In other words, this goes to show, you are being a subject of false sense of profitability. The results being good, does not mean anything. It can be a bunch of things. We don't know for sure. And that's the point.

We're uncertain. We need a measure for uncertainty.

This is where Timothy Master comes in. I have found his concepts to be sound and logical.

One of his concept is called "Selection bias" which occurs when the process of choosing the best model or trading system from among several competitors introduces an optimistic bias in its estimated future performance.

This is exactly, what my situation is currently. I have built a model. Okay, great. What now? How do I proceed with as much realism and criticism as possible.

He gives several workarounds to deal with Selection Bias.

•Employing Three Independent Datasets: To effectively eliminate both training bias (where a model learns noise as legitimate patterns) and selection bias, the author recommends using three independent datasets: a training set, a validation set, and a test set.
1. Training Set: Competing models are initially trained on this set7.
2. Validation Set: After training, unbiased estimates of each model's capability are computed by evaluating them on the validation set. The best-performing model from this set is then chosen. It's acknowledged that this chosen performance figure is still inflated by selection bias.
3. Test Set: The final step involves evaluating the selected model on a completely independent test set. This provides the ultimate, unbiased estimate of the model's true ability.

•Nested Cross Validation or Walk-Forward Testing: When the goal is to find an optimal predictor set or a model of optimal complexity, cross validation or walk-forward testing can be nested inside an outer loop to account for selection bias. While individual performance measures within this competition might be unbiased, the performance of the winning model is still compromised by selection bias. Embedding the selection loop within an outer cross validation or walk-forward loop helps to obtain an unbiased measure of the overall model-creation methodology.

•Monte Carlo Permutation Testing: Monte Carlo permutation testing can be extended to specifically handle selection bias. This method is used to assess the probability that good observed out-of-sample results could have arisen from random good luck. The "SelBias" program, for example, demonstrates selection bias by splitting a trading system into two independent parts (long and short positions) and generating two independent out-of-sample (OOS) datasets. The selection bias is then calculated as the difference between the winning model's performance on the first OOS dataset and its performance on the second, completely independent OOS dataset. This process is repeated multiple times, and the averaged quantities, along with a t-score for the selection bias, are reported. This test helps determine how often a chosen system is genuinely superior to its competitors versus simply being "lucky". The primary motivation for this algorithm is to accurately account for selection bias, preventing the confusion of unwarranted good luck with true predictive power.


-Source: Assessing and Improving Prediction and Classification Theory and Algorithms in C++ — Timothy Masters

These workarounds and many of his other methods, are really interesting and logical.

I thought that making a model is hard and gets the job done. Turns out, the hardest part, is testing it.

I just wanted to share this, because seems like some people in the forum are interested in Algorithmic Trading like krw.

I'll be applying many of his concepts to my use, as they seem to be the most rigorous and logical.

Cheers.

Solid insight. tnx