Serial vs. Parallel Processing in Asset Management Research

I put a poll out on LinkedIn on serial vs. parallel processing as it applies to asset management research workflows.

Two votes. That’s it. But I don’t think my audience understood the question well. Also, no one voted for “serial processing” even though that’s what most asset managers do.

Everyone knows parallel processing is faster because multiple jobs to be done are happening simultaneously. In serial processing, one job is done, followed by the second job, followed by the third job, and so on until the work is complete. Let’s dive in.

According to the new Wikipedia, “ahem” ChatGPT, these are the differences between serial and parallel and some pros/cons.

This is how an analyst workflow is a serial process:

A company reports earnings.

The analyst prints and reads the earnings release/slide deck and take notes.

Update model, revise forecasts.

Listen to the earnings call, take notes.

Write a formal “Note” to be distributed to colleagues.

Then on to the next earnings report.

This is a slow and methodical process. It requires force ranking which earnings report should be done next, and next, and next until the backlog is emptied.

Imagine Henry Ford had the idea to have a single person assemble an entire car. How fast would that be? We’ll that’s more or less what we do.

The only way to speed up a serial process up is to speed up the steps such as, have a better analyst that is better equipped. Maybe better at modeling/Excel, a faster reader, better at Bloomberg commands, quicker to understand the key points and messages in a press release or earnings call. The other way is to add tools to de-bottleneck the bottlenecks like a faster PC, FactSet Callstreet for quick takes or www.Daloopa.com or www.Canalyst.com for model updates and so on and so forth.

Like computers, parallel processing is much faster based on the specialization of labor. Only it’s hard and expensive. It also requires an insane level of trust at each process step.

This is a hypothetical example of parallel processing in an asset management workflow:

A specialized team prints/reads/reviews/summarizes the press releases earnings calls with the help of AI-powered app like Aiera (www.Aiera.com). This team only specializes in the crucial insights from the conference calls.

At the same time, a specialized model team, updates models with the help of an AI-powered app like Daloopa (www.daloopa.com) and then models out forecasts.

Finally, a content database with an AI-helper like www.DoTadda.com pulls this data together, looks back at the historical context, answers your questions and writes a draft note. Full disclaimer: DoTadda is my side project.

So, what method is best?

Well, it depends. For low-turnover funds I think a serial process is fine. For value funds, serial is fine too. I’m guessing Warren Buffett reads the Coca Cola transcript whenever it gets to the top of his pile. I doubt Warren is thinking about trimming some Coke during an earnings call.

For higher turnover and growth managers, parallel processing is the ideal, because the conclusions come faster.

For teams that generally want to process stock ideas faster, parallel is also the ideal.

Who is doing parallel processing today?

Wall Street research firms are doing it, sort of. Each senior analyst has a team of associates doing the mundane stuff like model updates and draft note writing and while the associate’s workflow is likely serial, there is at least a hybrid workflow going on here. All is this work is being done and finally the senior analyst signs off on the note and model and off it goes to the editor/compliance before being sent out to clients.

Big hedge funds that trade in and out of ideas quickly. I can’t speak from experience, but I assume a great degree of task specialization among the staff in an effort to quickly assess incremental data points and position accordingly, is in place. In this world, there can’t be a backlog of models and thesis updates. I imagine their are using all sorts of internally and externally built tech and services to move information through the process, into models, into notes, and ultimately trades. Maybe even with redundancies? The beauty of scale.

There is more to write on this topic, but it’s time for bed.

If you liked my thoughts (which are my own, not of my employer or former employers) please subscribe to my letter. I’m also on X.com @DrewMMeister and LinkedIn here: https://www.linkedin.com/in/andrew-meister-7254104?utm_source=share&utm_campaign=share_via&utm_content=profile&utm_medium=ios_app