Is there trading software that can identify and analyze similar trading days?

[ms33]

Let's say it's Thursday before the Payroll numbers at the beginnin of a new month. Is there software that can look back over several years and consolidate the price action for a given security coming into and out of the Payroll number? Be nice to have some stats before making a trade.

I know this is the despised nearest neighbor algorithm. I'd still like to see how a security has behaved in the past in specific settings.

Some practical constraints: 1) only compare the security with its own history; 2) sample the security's price action from the same time of day.

 

[ETJ]

Not a great answer, but there used to be a service that did this - you have to be over 60 and worked on Wall Street to remember. It was called ILM - Intelligent Learning Machine and the company was bought by Morningstar. They would even run your requested simulation by phone for a $$. It was mostly used for risk management and it had a huge hard drive(for that time period). It let you program in a series of variables like "if it's - the day after a holiday and the S & P closed lower last business day - how frequently does the dollar trade down?" Today it's mostly done by AI products - ILM was the only service I knew of and it's gone. Many discount firms had one for stress testing naked option positions.

 

[ms33]

Any idea what Morningstar did with ILM? I'll search that. Discount firms "had one for stress testing naked option positions." Is it embedded in a simulation or is it somehow accessible to a user?

Looks like Morninstar is using ILM to automate its quantitative rating system.

https://www.morningstar.com/blog/2018/03/05/machine-learning.html

 

[ZBZB]

Try https://markettells.com/welcome-to-markettells-com/ it does something similar to what you are looking for.

 

[ETJ]

As I understand it Morningstar killed ILM around 2000. I believe what they are doing now is running an in-house AI. If you had one you could pretty much simulate anything that they had a database for. Sat on the desk - a big desk - and you could run anything you wanted. There a handful of vendors pitching something similar today, but costs are phenomenal. Our desk is playing with one now and it's built on the IBM Watson backbone. This is part of the reason almost all of the trading firms have folks knowledgeable in AI and Machine Learning.

 

[Laissez Faire]

Create it yourself or have someone create it for you.

 

[Kevin Schmit]

It was called ILM - Intelligent Learning Machine and the company was bought by Morningstar.

Are you sure you are not thinking about LIM, not ILM? Logical Information Machines. Quite a lot of excitement about it in the early to mid-nineties. There was a competitor, QA (Quantitative Analytics) founded by ex-LIM NY salesman Bill A., bought out by Thomson Reuters. And a third similar firm bought out by CSFB, can't remember the name but was out of Boston, maybe "Tech Partners" or something similar. None of the three gained any traction after they were acquired -- kind of a pattern for this type of firm. Does anyone else remember NeuralWare and Casey K? Generated a lot of buzz in the industry early 90's, but ultimately fizzled out.

 

[Kevin Schmit]

[LIM] could pretty much simulate anything that they had a database for. Sat on the desk - a big desk - and you could run anything you wanted.

It was an error maximization machine. Allowed, nay encouraged, overfit. Lead to some truly disastrous trades at some pretty big firms. I, regrettably, have personal experience in exactly this.

There a handful of vendors pitching something similar today....

There sure are! Why do I expect that the ultimate outcome will be the same?

 

[tommcginnis]

Create it yourself or have someone create it for you.

I built one -- taking nothing more sophisticated than what's available for download off of Investing.com [OHLC and volume] data, and then sorting things into buckets.

I've dug this out -- apparently from 2016 as last update -- it's taken me 15 minutes just to remember how the thing flows together.
• The data sit on the left.
• the counting 'buckets' are set up across the top (columns H→W); they are symmetric.
• the conditional logic is set up in column G: samples appear in A1:A17
• Cells H17,H18 {their current rows} are the biggest single keys (See A1:A17)
• "Qualifying" dates are 're-printed' with their values (allowing a visual "logic check"!) from column G, for as many times [columns across] as they qualify.
• Counts ("@ISVALUE?? or some such) of the qualifying columns go on in row 22 + 23.

In the screenshot, there is 1 negative result and 3 positive results. For a lunchtime exercise, you should be able to discern (from these meesly data) what the operative conditional was.
[Whoops. I got curious, and brought the actual sheet up. Wayyyy off to the right (column AA) was a day-of-week column. The conditional (column G) is set to print only Friday results. Thus, 479 of 2428 market days were Fridays. 207 were net down days, 269 were net up days -- so you can figure the odds of a rising Friday right there...]


I found these handy when in tight spots on weekends or preceding FED announcements. But this was for an index. I would be curious to see the outcome for an individual equity. It would certainly provide perspective before earnings -- but echoing previous posts, LOTS of caution and a dose of salt would be wise.

 

[Turveyd]

I experimented just using screen shots, you quickly realise, there is no such thing as 2 days the same ever, more hassle than worth!!