Parker requested the information ratio for stocks a while ago. P123 has been working hard doing European data, DataMiner downloads that I particularly appreciate as well as the AI/ML that is coming soon.
Not to mention about 1,000 other things. Oh yeah, they had to switch from CompuStat as it was getting hare to justify the costs of the limited data CompuStat was will to provide at the time.
Understandably, P123 cannot provide every single metric we desire (including the IR).
But you can get the information ratio trivially now. The file is a simple download from P123 statistics for an ETF sim.:
I say this with no irony and only appreciation for my DataMiner downloads particularly, but also for European data and the upcoming AI/ML. We have the information ratio trivially available through P123 now!
Some like Spearman’s Rank correlation for the rank performance test. It was not super hard to do.
But probably not easy the first time for those who have not tried it yet (and wish to).
And it will be easier for me with this.
You could have it bootstrap returns also (again if you are interested).
All without having to make another feature request for either of those or additional metrics people may prefer. It gives us all some leeway without having to get P123 to pick priorities while working on bigger projects for us.
Excel has functions that can do this. Is there something wrong with those? t.inv., t.test., and t.dist. are the ones I’m thinking of. I haven’t used them myself so I’m curious why you preferred to use ChatGPT over the built-in Excel functions.
You need a one-sample t-rest or a paired t-test to do it right.
Not 100% sure Excel has those. But as I recall, the one-sample t-test can be calculated in Excel easily enough.
I’m not sure every member can do that off of the top of their head.
Just speaking for myself, we don’t mind a little help now and again.
ChatGPT can help.
I do appreciate for example that Dan helped with DataMiner when I just could not get it. Probably should have been able to I know.
But ChatGPT is what made ir possible for me to get the excess returns relative to the universe using group in Python to get the mean return for each week.
Not everyone can do everything right all the time on their own and not get stuck on something.
Thank you Dan and thank the forum for discussing some of the machine leaning basics.
ChatGPT helps me. It helped me with getting excess returns from the DataMiner downloads when I needed it…
TL;DR: ChatGPT can give you access to the Sklearn knowledge-base and coding now and without a feature request. It turns out that some of that is actually important if you want to do machine learning.
It’s not just about a ChatGPT making a t-test easier to do. That would be a distraction.
There is something seriously wrong if we are going to spend a lot of time at P123 figuring out the best way to calculate a t-test in Excel. If Marco is serious about bringing machine learning to the platform that is.
If Marco is serious about machine learning we should have been beyond that a long time ago. Maybe he thinks he can hit the ground running now. Probably he can, but he will need some help from people who take machine learning seriously to make it happen.
ChatGPT—which is knowledgable in the field, can code and has time to write and share code, and does not discourage people from discussing machine learning–is just one thing that is already helping Marco. Can there any doubt about that?
Maybe just me, but I think a major machine learning site (or someone who genuinely wants that to happen for P123) should not be having a debate about the best way to do a t-test in Excel and discouraging a discussion on how to make it easier.
It is good that P123 succeeds with this even if you love spreadsheets. P123 will be able to expand their services and you can make a feature request for your own ideas: one that will probably be heard. P123 will be in a position to fund your ideas if they succeed with machine learning. In that regard, I honestly do not get it. I think Marco has a noble purpose that should be supported by all of us. Marco certainly has the right to make machine learning part of his business plan if I understand the ownership structure at P123 correctly. You should help him beyond your EXCELLENT FEATURE ENGINEERING SKILLS that I use EVERY SINGLE DAY. Yuval, thank you for that. For developing some of those features, helping to code them with FactSet data and sharing many of your ideas in the forum. I am not trying to minimize your many strengths and your importance to P123 (obvious organizational skills being just one of many other strengths that I do not have the time or space to include here but area a clear benefit to P123). Seroulsly, I admire your organizational skills: the most important thing to being a good doctor AND GOOD TRADER, I think. Multiple check-lists can only go so far when you are disorganized genetically (talking about me here). Checklists I can never seem to find.
You might think about supporting machine learning at P123 is my only point.
I think you must have misunderstood or misconstrued my question. I was specifically asking Duckruck why he found it troublesome to calculate p-values and t-values for his data. His answer was quite illuminating. I did not mean to imply that people should figure out how to calculate a t-test in Excel rather than using ChatGPT or Python to do so. I just wondered what he found troublesome, that’s all.
My point was not about how to best calculate a t-test.
I do not think that I misunderstood that Marco wants to bring machine learning to the platform.
He said he is is committed recently in the forum. I believe he has purchased new servers, It has taken time and probably time means money with regard to the work of his machine learning expert and the hard work of programmers, His time and effort have value too. I believe him and I support his efforts. Am I wrong that he is serious about wanting machine learning to succeed at P123?
We are, in fact, talking about calculating t-scores and p-values in a spreadsheet at P123. That is a observable fact. Also we have talked about calculating the Information ratio in a spreadsheet, Spearman’s rank correlation, in a spreadsheet……There are few statistical metrics that have not been requested as features that, if you want to use them, you can….calculate it in a spreadsheet. Am I wrong about that?
Is that common for a site that calls itself a machine learning site?
Some people have real problems with the metrics (non duckruck who just finds it easier). It is now something ChatGPT might be able to help P123 members with. BTW, I actually made the point that these feature requests may be less needed which would be a positive point for P123.
More importantly, any machine learning method will require metrics that are………(you guessed it) not calculated in a spreadsheet. I think that is true, right?
ChatGPT just happens to be available now. There is nothing P123 did to make these metrics more available or easier for members.
Riccardo said this in the forum:
I do not mean to put words into you mouth. Maybe you can tell us which of these you are exited about.
I appreciate you were being humble with this and I actually appreciate it:
But I am guessing regression might not be one of the methods you support and some of what I have said is based on that. So is it neural nets you will support–if it is not regression–when P123 releases its AI/ML? Support Vector Machines, perhaps?
My post was not about how to do t-tests. And I do not question Marco’s commitment. Just tell us you support machine learning at P123 and we can move on.
Jim, I confess I’m completely confused by your response. I asked Duckruck what he found troublesome about calculating t-tests. You then said that there’s something seriously wrong if we’re going to spend a lot of time figuring this out. Well, I didn’t spend a lot of time figuring this out, and nobody else did either. I just asked Duckruck what the problem was. You then started writing about machine learning, which as far as I know has nothing to do with t-tests, except that machine learning can use t-tests. Machine learning wasn’t even conceived until fifty years after t-tests were first published. Then you seem to imply that if P123 is going to do machine learning we shouldn’t be talking about spreadsheets. That’s like saying that if Microsoft is going to be doing generative AI they should get out of the word-processing business. Lastly you ask me to tell you that I support machine learning at P123. Of course I do. I’m very excited about its prospects and am definitely interested in developing new factors using it. Who wouldn’t be interested in checking out a new tool? But why should it matter to you what I think of machine learning? And what does it have to do with t-tests and Excel?
Who would you put in charge of supporting and explaining machine learning in the forum if you were heavily invested in the hope of people try machine learning, ultimately accept it and further developing it with thru feature requests in the forum?
Me personally, I would want that person to understand why Riccardo is excited about bringing regression to P123. I am just going to say I think that would be important to me.
It would be a deal-breaker for me in any job interview to be sure. Maybe my business is different but I would not want one of my employees telling a patient she did not understand why I was recommending cataract surgery.
I also said Marco has made a serious investment in machine learning working for P123. You understand that part don’t you?
I still don’t understand. Who said I was in charge of explaining machine learning in the forum? I’m the last person you’d ask to do that. I know much less about machine learning than Marco or Riccardo. We all have different roles here, and machine learning isn’t one of mine. Why go on the warpath against my innocent questions? I would love to understand how PCA analysis can be used to actually choose stocks to buy and sell, and in response to that question someone responded with some links to some valuable material. Am I not allowed to ask questions like these on the forum because I don’t yet understand machine learning?
Of course I understand why Portfolio123 is recommending machine learning. It has enormous potential and I know exactly what that potential consists of. I could explain that potential to someone who knew nothing about it. Would you fire your employee if she didn’t know how to perform cataract surgery?
Also, FWIW, I’m not employed by Portfolio123. I retired. I still do a lot of work for the company, but unpaid.
My employee would understand what a cataract is (and why I do them) before I gave them any responsibility for patient education.
Call me demanding on that count.
You are supposed to help with product development? Answer feature requests? Maybe answer questions about a method in the forum? Who is going to answer questions in the forum about machine learning at P123?.
Tell me you have done one support vector machine. Just one? At least one would be good if you are going to be the person from P123 answering questions.
I am just saying Marco needs someone else in the forum that understands this and could answer a feature request with some understanding of what if being asked.
Even if you you have suddenly, inexplicably, become the new champion of machine learning in the forum you would have to understand it. Or P123 would have to hire someone who does understand it.
TL;DR: You do not understand machine learning methods. Marco should consider hiring someone who does if he has not already. I think he will need to if he does not want people to just use the API. It is possible that just selling downloads will work for him but then why all the new servers?
