Auction Market Value Theory & Analytics

A quick disclaimer first:

THE READER SHOULD BE AWARE THAT THE BACK-TESTING IS DONE USING MY OWN METHOD OF CLASSIFYING MARKET CONDITIONS. In other words, I AM NOT USING THE ODC THAT OTHER MEMEBERS ON THIS FORUM ARE USING. WHY IS THIS DISCLAIMER IMPORTANT, YOU MAY ASK? IT'S IMPORTANT TO SAY THIS, BECAUSE YOU CAN'T REPLICATE MY RESULTS USING THE TRADITIONAL APPROACH TOWARDS CONSTRUCTING AN ODC, YOU WILL NEED TO BUILD THE ODC IN THE WAY I HAVE DESCRIBED, WHICH IS SOMETHING THAT'S MAJORITY OF THE MEMEBERS AREN'T WILLING TO DO. HENCE, DON'T BE FOOLED THAT THESE RESULTS WILL BE THE SAME FOR YOU ASWELL, BECAUSE IT WON'T BE.


IF you haven't read the disclaimer, kindly divert your attention to it. It's important that people are not mislead.

So, the topic for this post is a basic stats and overview for the Back-Test 2024-2023.

The excel files are attached at the bottom of the post. The names, contents and important notice about the files are as follows:
File Name: "Back-Test Results"
Content: The result for each trade taken during 2024 and 2023 period. There are metrics that have been explained before-hand, however, there were some important context left. The context is as follows:-

1. There are two types of Strategies/Risk Management Methods. The STR TYPE 2 trade's risk management assumption is 20 Day Volatility based SL by 2.5x. Fixed TP of $5. Micro Lot of 0.01. STR TYPE 1 was described in an earlier post.
2. Trades that don't reach TP or SL, are assumed to be closed at the exit price. These are labelled as EOD. Their P/L is monitored for the next day.
3. The last column named "SS", included Screenshots for each trade. However, those have been removed for 2 main reasons. One, the file was becoming unstable as there were 200 SS and Secondly, there's a risk of security for the user and for me. Hence, it was removed.
4. The News column has also been removed. The reason is because there seems to be a subjective opinion regarding what is considered a "High Impact News Event" and "Moderate News Event". To further complicate the situation, the news data for 2023 was lost, hence, it has been removed altogether.
5. You will see duplicate dates with different Trade ID's. This is intentional. Maximum trades per day are 2. The only the model is allowed to trade twice is if the two conditions are met. 1. The SL Price was hit. 2. Entry Price was revisited.
The whole period of 2023 contained many trades that happened twice. Hence, if you are not willing to perform this as a strategical tactic, it's wise to remove them from the analysis and the final result.
6. Note! The results are based on a Chronological Distribution Sorting approach. Meaning, the results are not meant to replicate yours.

File Name: "Bracket-Screen"
Content: Contains market condition for each typical timeframe, 1 and 3D VTY, as well as checkboxes that indicate when a trade was executed.

File Name: "GBP-USD Flow-Analysis & RI-QC 2024-2023"
Content: It contains Flow Analysis for each typical timeframe, RI-QC till 2024 and 10 custom calculations whose formula's are left in the workbook for the reader to explore.

None of the file contains macros. These are read-only files.

Great. Now, that we have established some context, let's get to the fun part. Analyzing the performance of the trades.

The results for the trades are based on STR TYPE 2: VTY Based Stop and Fixed TP.

_______________________________
Metric Value

Total Trades: 193

Winning Trades: 90

Losing Trades: 103

Break-even Trades: 0

Win Rate: 46.63%

Loss Rate: 53.37%

Average Win: $3.56

Average Loss: –$2.53
Max Drawdown: –$34.19

Profit Factor: 1.23

Expectancy per Trade: $0.31

Net Profit: $59.30
______________________________

Interpretation:

Win/Loss Structure

1. Despite having a sub-50% win rate, average win is significantly larger than the average loss, resulting in a positive expectancy.
2. The Profit Factor of 1.23 suggests acceptable profitability, especially for a mechanical, no-discretion system.

Expectancy

1. Trade's are making ~31 cents per trade on average, which is positive, but modest. Expectancy could increase dramatically with flow analysis overlay or better trade filtering.

Drawdown

A max drawdown of –$34.19 on a net profit of $59.30 yields a Return over Max Drawdown (ROMAD) of ~1.73 — not bad for a mechanical system, and quite respectable for micro lot size back-testing.

These were some of the basic insights/metrics tracked in almost every historical back-test presents. The only other things I didn't include were Sharpe Ratio or other ratios. Which honestly, aren't of much use.

However, we have tracked a lot of variables, hence, we can tune our understanding by looking more deeply into the data. For instance, our next steps for a deeper analysis will include:

1. Analyze MFE/MAE behavior to understand reward-to-risk skewness and potential premature exits.
2. Assess TP Hit After SL Hit? — this might show tight SL issues and opportunities for dynamic or volatility-adjusted SL.
3. Evaluate Entry Price Revisits and Trade Continuations for deeper behavioral insights.
4. Generate time-in-trade performance (efficiency) and possibly a few performance visuals.

Observe the Graph Below:

Metric Value

Average MFE (Profit Potential) 4.10 pips

Average MAE (Drawdown Before Exit) –3.67 pips

Chart Observations:

1. The MAE distribution is slightly tighter and more centered around small losses, which shows that most losses are modest.
2. MFE, while more dispersed, has a long tail, meaning some trades had large unrealized profits — likely not fully captured before closing. This is the biggest edge to exploit.

This edge is exactly what lead me to create dynamic TP, which gave results much better than STR TYPE 2. It's just waiting to be explored by members who are willing to do what it takes..

I will leave other 3 steps unfinished, so that members of this forum are encouraged to take a route and contribute to the forum.

Conclusion:

Were the results satisfactory? Absolutely! The results could not have been better, even with a very standard risk management technique, the results are better than expected. The most fascinating part out of all of this, is the ability to code this as a mechanical strategy, without including any sort of analysis, and still be in Net Profit.

What does this tell us? This emphasizes the importance of Market Condition. By just trading based on Dynamic Market Regimes, we are able to exploit market phases, thus, market prices.

Is the AMVT Model a Money Making Machine? A STRONG NO! Albeit, this data is statistically significant, it can not take into account what the future holds. Hence, this can't and shouldn't be treated as blind system.

The results are convincing that it's a good mechanical system, so what makes you say "A STRONG NO"? The reason the "Model" performs well is because it's Dynamically Updating Market Regimes. If the Market Regimes become unstable, for instance, there's still no way of Quantifying A Choppy Market, then the results will be way different. Even though, the "Model" will identify it as balanced/bracketing phase. Hence, caution and some common logic should be applied before jumping the gun.

What's Next? The post after this will explain in detail what the next steps should be or can be. There's still a lot of work left undone. Hence, we'll talk it through in an essay type of format.

P.S. Your responses, suggestions and contribution is very much appreciated. Hence, I'll be waiting for your remarks.

Cheers.

Importance Of Data Validity

One shouldn't trust data, even if it's taken from the market itself. The reason is simple, it can contain mistakes, errors, inconsistencies, noise and all sorts of formatting and other issues that causes issues with data validity.

Hence, to exercise precaution, it's necessary to acknowledge this fact, so one can stay protected from unrealistic expectations and false results.

I am talking about this for a specific reason. The files I have shared, may or may not contain data that is fully correct. At the end of the day, I'm still a human, and humans makes mistakes. Hence, if you are planning to do any sort of planning with the data, i.e. applying it into an Algo, using it for your own personal analysis, or any other practical application, then make sure to double-check everything. If you are to find inconsistencies/errors in the data, it's important to let us know, so I'm able to upload correct versions of it, if necessary. And people are refrained from using earlier version.


Discouraging Misuse

The results, data, analysis, and files, should not be treated in the intention of making more money. I have shared these files, for the sole purpose of encouragement among members of the forum to pursue research and join in this venture.

We are able to help each other out, and the only way we are able to do that, is if we are to become self-less. Hence, if I have been any use for you, then you are owed to the forum to contribute any insight, information, improvement you come across. Of course, protect whatever is not worthy of sharing, however, contribute as much as you can, so we can learn from each other, grow with each other.

That is what makes a community.



Next Steps For Me and You

Everybody will have their own plans. Some are going to ignore the back-test results altogether. Some will take interest. Some will take interest but not participate. Some will be silent observers.

For those who are serious. I invite you to an exciting journey ahead.

I hope whomever these people will fortunately be, will post their findings here. So everyone else is also benefitted and encouraged.

My personal plans includes the following:

I will be applying Topological Data Analysis (TDA) to the flow, back-test results, RI-QC, and other custom calculations.

TDA can help figure out patterns, anomalies and understand the structure of the data on a level that is hard to achieve without it!

Anybody else interested, can go to the Sci-Kit TDA website to learn more about it. Their website has detail documentation, steps to apply this in python and general information regarding TDA and it's associated methods like Keppler's Mapper & Persistent Homology. Don't get scared by the names, the steps to apply TDA is as easy as basic print hello world code. Max 50 lines of code. Easily done via using an AI, some education on TDA and it's associated Python libraries or a plus if you already know programming.

For people who have suggestions on what else I should do, kindly let me know. I am very much intrigued to hear new and fresh ideas.

Hopefully the forum will become vibrant again. As that was my goal from the first post I did, ever.

See you soon with more updates.

Thanks for reply. But I came to some conclusion, which I dont understand now. Its going against cisco methodology. Please AMVT, KRW or JP answer so I can see you thoughts and opinion on it. We can see on SS, how brackets limits are down but price never got to that limit. How is that possible, we can only see volume there. But for example 3. bracket box we can see how price never went down to those levels where bracket LL is. Therefore I dont understand. In resources it is said that bracket limits are determined by TPO counts. Anything above 3days we use 5tpo count....Please explain, maybe I am just a newbie. Thanks in advance

Can you also reply to my private message. Thanks

For your information, this is today's GBPUSD

The amount of things you said that were absolutely wrong honestly goes to show how limited knowledge you have for AMVT.

First of WHAT VOLUME ARE YOU TALKING ABOUT?? The ODC is made up of TPO's (Time Price Opportunity/That Price Occurred), which counts the frequency of ticks/price. TPO's are the smallest unit of measurement. It does not represent Volume!

If you don't even know that, then please refrain from using AMVT and get to know the basic concepts, first, before applying it.

You are viewing the market in a 30 minute TF. Hence, the low that made 5 days ago, or the lower bracket limit that made 5 or even 3 days ago, will of course not be visible in such a small TF. You are viewing a short time space, where-as ODC represents larger/longer TF's.

The indicator only rents space on the chart, it doesn't represent the candle-stick prices exactly.

What you can do is see the dates of 3 day. Zoom out, and get those dates in your vision. That way you'll be able to "see" when those prices were traded.

I'm repeating the same exact things MzVega did 6-7 years ago! If only you care to read about it in the forum, all your trivial problems have an answer. You have to look for it.

No. Any TF above or equal to 10 days, 5 TPO count is used. It depends on your own personal choice as well. But that's the "guidelines", if I remember correctly.

I advise you to read the thread. Spend just one day exploring the thread. Spend another day reading the necessary material. All your questions will be answered. But care to read.

Hello Members!

Don't see any activity so I though I might come in with a bit of an update.

As previously stated in my Back-Test Overview Post, I am applying TDA to the back-test results, in the hopes of finding patterns, anomalies, interplay among the variables etc.

The process has not gone smoothly, but nevertheless, I have finally completed TDA Data-Set as well.

To motivate people on doing some research, I will share some results I have gotten through Bcak-Test. Don't get scared by big/scientific names. Everything will be explained thoroughly in a not so scientific way to help convey the importance of TDA and in general doing research.

Some Background on TDA First:

Topological Data Analysis (TDA), is a relatively new and emerging field. It helps to find the "Topology" of your data-set. In other words, it helps to see/understand the Structure of the data. It doesn't care about the values of your data-set. Rather, it looks for the Geometry, nothing else.

There are 2 Popular Methods in TDA known as Persistent Homology & Mapper Algorithm. Both are very easy to understand if you give me a chance to explain.

1. Persistent Homology: Imagine you have 2 data-points in a 2D environment. You grow Balls of some radius you determine, around the data-points. When these balls grow and grow, the 2 Data Points will eventually be overlapping, in other words, The Balls' radius on 1 data-point, overlaps with 2 data point.
See SS:

The "Small Dots" are your data points. The circles are our "Balls". Right now, they are not overlapping.
See SS:

Now the overlap is complete. In a nutshell, that is basically Persistent Homology!

Now, let's dive a bit deeper. First I want you to logically Understand Persistent Homology (PH) by their literal meaning. "Persistent" something that is determined, robust, doesn't fade away. "Homology" another word for holes/cavities.

Now, how does the literal meaning fit into our nutshell statement? And what does this have to do with TDA??

Basically, you're tracking the "death" and "birth" of your data-points when the balls are growing around.

In other words, you're tracking, when the birth of a data point occurs, this will be very early on, and the death depends when the balls radius overlap with each other. So, when you track that, you get a diagram!
See SS:


When you track the "birth" and "death" of your data-points. You are able to create the necessary geometry to understand the shape of your data.

A good question is: Why understanding the shape of the data helps? What information does it provide? How can it be practically useful?

Why Understanding the Shape of Data Helps:
1. Limitations of Traditional Methods: Traditional data analysis techniques often focus on pairwise similarities (like Euclidean distance) and may disregard crucial information about shapes, including loops and holes. Many algorithms also rely on simplistic assumptions about data shape (e.g., assuming spherical clusters in k-means). This can lead traditional methods to fail at finding significant patterns, such as spatially distant points that nonetheless exhibit similar structure or shape.

2. Revealing Hidden Structures: Data, especially complex, high-dimensional data, contains intrinsic structures that are not easily discernible through conventional methods1.... TDA provides a framework to uncover these hidden patterns and structures. High-dimensional data is often sparse and tends to have relevant low-dimensional features, which TDA can characterize precisely.

3. Robustness to Noise and Metric Choice: TDA offers methods robust to noise and missing data.... It focuses on the arrangement and proximity of points and how they are connected, rather than relying solely on specific numeric attributes or the particular metric chosen.... This invariance to specific geometric transformations means that the skew, size, or orientation of data doesn't fundamentally change the core topological insights.

4. Providing a Unique Perspective: Understanding data shape through TDA offers a fresh perspective on data analysis, enabling the capture and understanding of qualitative relationships. It allows researchers to see beyond the surface of data. Datasets with identical summary statistics can possess very different shapes, and visualizing these shapes allows analysts to gain far greater insight than relying on summaries alone.


I can list many, many, many more points. But I hope you get the idea.

How Understanding Data Shape Can Be Practically Useful:

•Financial Analysis and Forecasting:
◦Understanding topological patterns in financial data provides insights into relationships, helps in problem-solving, risk mitigation, and data-driven decision-making.
◦TDA captures intricate and multi-scale financial data structures, offering insights into market behavior and predicting critical events like market crashes.
◦Comparing persistence diagrams or their summaries (e.g., using Wasserstein distance) can quantify structural divergence or similarity between segments of financial time series data, helping identify significant market shifts or anomalies.
◦Ball Mapper can identify regions in characteristic space where specific outcomes occur, such as firms failing.
◦Topological features can enhance computational forecasting models and quantitative finance strategies.

If you don't believe me, Google it. You can find articles in Research Gate, Archivx.org, and hundreds of other publication websites that provide empirical evidence, from scientists around the world.


So now that I have (hopefully) helped you understand what TDA and specifically PH is about. I will now proceed to show the interesting results I got.



H1 Persistence Mean: Average lifetime of loops. A high mean persistence might indicate strong cyclical patterns, while low means may suggest noise.
This metric is related to the Persistence we talked about earlier. However, the method that I am doing, is called a "Univariate Time Series" which is a rolling window method. Hence, we apply the PH at specific chunks of the data-set. For instance, here I have chosen the rolling window to be 20 days. So, every line represents Persistence mean calculated from 20 days of data.

As you can also observe instantly, there are many spikes. Indicating the GBP-USD market is going through strong regime changes. This can be very helpful, especially in context of Flow Variables and understanding of Market Condition.

Another metric:



H1 Persistence Max: Longest-living 1D hole. Indicates the dominant topological structure in that window.

As you can observe, there was a huge spike, which indicated, something structural happened, deep within your data-set, it changed the whole Topology of your data-set.

There are many, many metrics you can calculate. Some will require the use of other TDA libraries, some are easily done in Sci-Kit.

The full metrics are shown below.

(The data set is till 500 rows, hence every value after that is flat)

Anyways, I hope this might spark an interesting debate in the forum. And I hope people are encouraged to take the challenge. As this is something not many people are paying attention to. Hope you all the best.

Next Steps

Filter Topological Features by Trades win/loss. Apply Mapper Algorithm to find clusters, that may lead to topological patterns.

Simple.

Thanks, Ill consider going through literature once again.

Krw could you please send me private message. I would like to discuss something with you. Thanks in advance

So far, I am completely amazed by how Keppler Mapper Algorithm operates, and its various practical applications.

I have completed building the Keppler Visualization, and first off, let me give proper background on it:

2. Keppler Mapper: It's the second-most used technique in TDA. It helps to visualize clusters through filter(s) of your own choice. In the context of TDA, this works best with topological data/features. Though, there any many other methods available for clustering as well, i.e. K-Means, Simple Scatter Plots etc. It's your personal choice. Even in Mapper, you have the choice to select DBSCAN, OPTICS or other clustering methods you want. It has a lot of area for experimentation and creativity. I won't get into the details. For instance, building Mapper requires dimensional reduction as-well. For those interested, I used PCA (Principle Component Analysis), this choice is also left for the user to explore.

I wanted to share the .html file for Keppler, which is an Inter-Active Visualization. However, there's a risk of security & privacy for me, hence, I'll refrain from making such links public. However, I can share a static image, that can help you understand what results I have gotten so far.

SS:

Let me explain what you are looking at:

1. The Clusters are the Colored Nodes you're seeing in the center of the screen.
2. Each color corresponds to a win/loss.
3. Purple = High Loss , Yellow = High Wins , Light Green = Wins mixed with some losses , Blue = Mostly Breakeven, or losses.
4. In other words, their percentage can also be used to estimate the wins and losses in each node/cluster.

Interpretation:

The fact that the Yellow nodes don't have any clusters, mean these are High-Probability Set-ups. Their frequency is low, meaning they won't occur often. However, when they do, you have a confident trade. Light Green nodes are also worthy of a look. Their win rate is 64% (for context, win-rate for yellow nodes is 100%). What's even more fascinating about this, is that Losses (Purple Nodes) have formed a cluster. This usually, doesn't happen. In a typical clustering output, you'll either have outliers like I have with Yellow nodes, or a winning cluster. A cluster for loosing trades is the most beneficial cluster you can get. It tells you when NOT to trade , as well as highlighting high-probability set-ups. Meaning, if you are able to only avoid these specific market conditions, you'll have a good year.

Conclusions:

The results are clear: AMVT operates in subtleties, not A+B = Win. Interpreting these subtleties in the trading data, honestly goes to show how much of a skilled trader MzVega was or is. In other words, it's not easy to catch these non-linearities. The interplay of variables among different timeframes is masked so deep, that, it becomes impossible by people/retail traders who are only focusing on the profitable side of a system.

For now, these results show promise, however, we still have the Risk of Overfitting. Hence, to make the results more robust, we have to experiment with different time windows for the PH Diagrams in order to see, if the results/clustering is consistent. The last thing left, is then, to test the findings in out-of-sample-data.
Only the patterns found, will be tested. If the patterns show consistency, we have a found a winning formula. If not, then this means TDA also has to be dynamically updated, like the Market Conditions.

Still, there's soo much experimentation to do. For instance, the Mapper I have built uses my wins and losses as filter, we can even filter by Market Conditions. This will lead us to create a highly sophisticated Hidden Markov Model (HMM). As you can see, the potential of practical applicability is vast. Not only practical applications, but as a data scientist, this is what I thrive in. And it's so exciting to be able to get these results and share it with members. Even if I don't get responses, I know somebody appreciates the effort and use behind this.

For now, my time is over in this forum. I have done what I needed to. That was, to show the people, that with research , you are able to do what majority only thinks is possible in a fantasy land.

I will be posting, only occasionally. but as of now, I don't see any point posting here, as no members have responded, which leads me to believe there's no interest left. But, I have delivered what I promised. I can't hand you everything in a plate, but I have showed you a path, that is different, and is worth putting effort in.

Wish you all the best. <3

AMVT,
Thank you for posting your findings. Given that you appear to be applying your scientific expertise to the data analysis, some of us may be hesitant to respond. The reason for our hesitation may be that the information is too complex for us, rather than a lack of interest. Therefore, I strongly encourage you to continue sharing your information, as we may understand it better in later times.
JP

Hello JP,
My only savior

I can't put into words how much I appreciate your encouragement and acknowledgement for my work.

You have been by my side since I started and I admire your loyalty.

I see the point you're making. It's very much understandable, that these concepts are complex. There's no denying in that.

However, don't assume that I am an expert in this as-well, and I say that for a particular reason. In fact, I learned all of this the previous week.

Allow me to elaborate.

When you encounter something new, a concept, a strategy, a new field, which is very often in trading. You don't start with the complexity. You start with, applications. That is done, so you are not overwhelmed by the technical details. Now that you're aware of the applications, you can delve into the foundations.

I have only learned TDA through books and youtube videos. And learned only those components of TDA that are helpful to me specifically, in my context.

I don't understand the formulas, I have never been good at mathematics. But I have a will to do something. I want to learn. I want to explore. I am not a smart person. In fact, I have made dumb mistakes during the whole process. It was frustrating. It took me 1 whole year to get only started with trading, not even talking about profitability. Meaning for 1 year, I was only reading, researching and testing.

I never knew how to code, I learned it during trading. I didn't know how to use excel, I learned it. I didn't know how to use/install MT4 indicators, I learned it. I didn't know what the indicators are or do, I learned it.

The point is, your IQ or the level of expertise don't matter when you're learning something new. You don't have to be perfect from the start. You NEED to do the mistakes, in order to learn. You need the practice.

So, if it's complex, ASK. Ask me, and that will not only help me consume more knowledge, but also enable the ability to share it.

The questions can be absolutely dumb, stupid. It doesn't matter, and I won't care. Because you're starting out fresh. But don't let the fear control your knowledge and potential.

However, even if I share the results and findings, if people are not motivated to learn, there's absolutely no use of the results. Because I won't be providing you with patterns, I won't be giving you the code. I can only help you in shining light to the path that led me to success.

Just like MzVega, or Evak, can't make you trade for you, they can only teach you. By you I mean all memebers of the forum. This message is not only directed to JP.

In fact, I belive strongly, that the only members who are capable to apply this are Senqiua and JP.

That's not to say that they are more educated, or have a higher IQ, they know coding, or they are great mathematicians. Rather they have the will.

@amvt

Your last post is interesting and I’m trying to think about your findings in comparison of my own results. I had built for me an EA that trades as an Alert Only or as an Algo. It is meant for trading futures using CQG data on MT5.

Now I’ve been backtesting for months on years of real tick and volume data. I am in the camp of Daniel Fernandez, of Asirikuy, who believes all past data is OOS. The only way to not overfit is to create a system with as few variables as possible, and test as far back into history as possible. The man who has done that is Martin Armstrong, but that isn’t a topic concerning Auction Market Theory.

Yes there is different interplay between TFF, VTY, VOL, VA, etc., but my algo testing results have been positive. I said before that a human would have better results than a hot/ cold algorithm, but I continue the fight regardless.

Want to learn something new? Google Notebook LM - Upload books, websites, youtube videos. Ask questions, generate a podcast. FREE.

Want help in coding? Claude & ChatGPT - Free versions are more then enough.

Want to code? Google Collab - Use Google servers, ask questions to Gemini for your code, help it fix your issues, no need to paste all your code, it knows your code if it's in Collab. FREE.

Want to explore topics? ChatGPT - Great for ideas, exploration, insights. An All Rounder.

Want specific code examples/documentation? Github - Use community's examples to train your A.I to understand the structure of the library you use.

Want to download books? Anna's Archive .org - 1 Billion plus pieces of written content. Books, articles, you name it. FREE.

I have used all of these resources mentioned above to get to where I am. The only thing you need to do, is to get started.

Hello KRW,

I appreciate your comments.

I don't have any knowledge of the camp you mentioned. However, I do know Martin Armstrong. Due to his actions and a bad public image, I refrain from talking about him.

However, your comments on overfitting are intriguing nonetheless. I don't understand the context behind "variables", please elaborate on that. However, assuming you mean variables in the context of market reference points/metrics like volume, ohlcv, tff etc, then I agree and disagree on that statement.

It depends on how you use these variables. For instance, there are a lot of moving parts in an Algo. For some components, you need each and every variable possible, for some, having not too many is ideal. But again, it depends in what you are achieving to do.

However, I don't think that having less variables leads to to not overfitting data. I do believe that testing as far back as possible, definitely helps with not overfitting.

Though, it's not as easy as it looks either. Having less variables = less overfitting, can work in ideal situations. For the most part, you need all of them, and afterwards, drop the redundant ones through testing measures i.e. correlation analysis, random forest classifiers etc. For instance, nobody knows what important metrics should be used. We have to know what the results will be, if we include all at once. And slowly, dropping metrics if they don't prove to be useful.

There's a method to do this. But it's not straightforward. Again, it depends on what you are doing.

In the context of TDA, these results can be overfitted. One test is to see whether these patterns occur in different time windows. Using the same data-set.
Hence, that will also give a rough idea if the pattern is robust or not.

But, either way, testing is important. No matter what.

Variables in a system would be categorized like choices between 1, 2, 3, 5, 10, etc for example. They could be numbers or other conditions based from a single input.

in the case of AMVT yes there are quite a few conditions, and they do mix and mingle, but I wouldn’t consider them variables. My approach with continuous backtesting has been to systematically align certain reference points for trade/ no trade set ups. These are hard coded and non negotiable, I don’t consider them over fitting. I did include other variables that may be suspect though. Still working on things….

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.

In theory, this should work. I have never built an algo, but it seems logical, that the market you're trading is not static, then the trade set-ups should also not be static. It continuously needs to find new or improved patterns, based on what method you use.

Either way, wish you best of luck in your project

the trade set ups are not static my friend, they are based on AMVT. They are not based on holism

That's perfect. Based on you're statements: "These are hard coded and non negotiable". It was reasonable to assume that you are making decisions based on if x+y happens = trade. Which is hard coded. If they're not, then it's good for your system. My friend.