Predictive Analytics: Machine Learning fit for Financial Modeling?

TLDR;

"Predictive analytics" refers to a group of machine learning techniques. They are used to determine quantitative relationships between observable variables. These allow us to make better forecasts and “what if” predictions.

I’ll argue that this is the part of machine learning that is relevant to financial modeling. Us modelers are already in the business of helping people make better decisions by building them driver-based business models. Predictive analytics presents a promising opportunity for us. We can use its techniques to create better forecasts of the underlying assumptions that drive our models.

That upgrades our models into even more powerful, nuanced, and accurate tools than they are today. There’s no reason why we shouldn't expect financial modeling specialists to have these skills in the future, in addition to the ones they already possess.

I’ll follow on with a review of an online course for predictive analytics provided by Udacity.

Disclaimer: the opinions submitted in this article on all products and services mentioned, are entirely my own. No discussions or arrangements have been made regarding their promotion.

Learning: the long and winding road...

I count my blessings that 2020 has not been as disruptive for me as it has been for others. I’ve been freelancing remotely since late 2017… so if anything, I’m excited about the rest of the world catching up to seeing remote hiring as something normal!

But I’m not immune to feeling *some* apprehension about the shakeup I believe is coming up ahead… So, like everyone else I have taken the cue to upgrade my skills!

My financial modeling is already high-level game, as is my VBA. So where should I look?

It’s pretty clear that there’s a strong buzz around analytics and data science.

These practices are going to create some exciting and disruptive developments for financial modeling (which has otherwise changed incredibly little since the 90s if not earlier). This year, I had to learn M-language and DAX to deliver some client work, and I even got an enquiry about training up a data science team in financial modeling.

There’s a lot to say about what is happening in this cross-over and why… but I’ll save that for another article (or two!)

So, off I went to Google looking for something data science/analytics related…

One option jumped out at me: “Predictive Analytics for Business”

I had the same reaction any honest nerd would tell you they had:

“Whoah! That sounds cool…!”

Erm… so what is “Predictive Analytics”?

With the previous, key item of diligence out of the way, I could dig a little deeper…

Predictive analytics comprises a subset of machine-learning techniques. Typically, they are models that are focused on forecasting, and they have a causal flavour (like financial models do). By this, I mean that they are used to quantify dependencies and correlations between things that you are able to observe. (Better still, between one or more things that you are able to control!)

Contrast this with automatic image recognition, which uses unsupervised machine learning methods. The goal of that application, is to develop a closed solution that eliminates human involvement. There is no primary concern with what relationships the model uses to make its conclusions, the only thing that matters is that it works!

Now let’s look at financial modeling. Modelers like myself, create tools to provide clients with a comprehensive and integrated overview of their businesses of interest. This empowers them to take investment and managerial decisions that have cross-departmental implications.

The goal of my new training isn’t to make a completely new career pivot… but to find ways to leverage what I already have! With this goal in mind, we can see that it is predictive analytics (rather than a machine learning course with a stronger AI slant) that offers a closer fit with the concerns I already work to serve.

Let me explain why.

While AI solutions can bring near-complete automation to select business processes, they are very far from being able to automate the full range of management actions and decisions. (Listen to those sighs of relief!) Humans are still needed to react to information and manage other humans to get things done. 

In this broader domain of managerial and strategic concern, the gains machine learning currently offers us are an improved capacity to process information, and to make astute decisions with it. This lines up with the focus we find in predictive analytics.

So how would predictive analytics improve my service offering?

These skills can put me in a position where, not only am I able to build a “classic” financial model of the business as a whole… but I will also be able to use these statistical techniques to create the models’ driving assumptions from real-world data.

In simpler cases, there wouldn’t be a need to hire additional consultants to perform this extra task. (Recall that most data scientists and data analysts DON’T understand accounting, financial theory, and Excel engineering to be able to model the way we do!)

OK, so predictive analytics not only sounds cool, it also makes sense as an incremental development for my current skillset. Now, which course should I take…?

Choices, choices, choices...!

The table above shows a snapshot of reputable courses with significant portions on predictive analytics.

Most of the courses are “expensive” relative to the expectations many of us have for online courses these days, and serious institutions are taking part.

You should see that as a great thing!

It means you are investing in a high value niche! If you’re approaching this as a modeler with some years of experience, the prices are far from being outrageous as an investment.

I chose Udacity’s offering.

They’d been on my radar for a while. They communicate a convincingly professional brand with their courses. They cite partnerships with high profile business and tech partners, and benefit from credible and charismatic leadership provided by Thrun.

The particulars that swayed me:

• Self-paced time frame (to make better use of time in between freelancing gigs)
• A broader introduction to useful tools including SQL, Tableau, and Alteryx
• Free 6-month licences for Tableau and Alteryx
• Fully downloadable materials
• Career mentor access & support
• And their 40% Covid discount!

What was it like?

I'm in the business of corporate/online training myself, so I know how challenging it is to build a good course. I took some notes from the experience!

The videos were professionally produced with good instructor delivery and meaningful and supportive illustrations and animations. They are tidily segmented into coherent, bite-sized units.

The topics covered are listed below. All were taught and implemented using the software tool Alteryx:

• Data wrangling (replacing, parsing, joins & unions, fuzzy matching, spatial blending, reshaping & aggregation, handling outliers & missing data)
• Linear regression
• Classification models (logit/probit; decision trees; random forests; boosted tree models)
• A/B testing (random test designs and matched pair approaches)
• Time series forecasting (ETS and ARIMA models)
• Dimensional reduction using principal components analysis
• Segmentation using k-means clustering
• Career coaching (tips, LinkedIn CV review, and mentor access)
• Tableau (and additional Tableau training offered as optional extension)
• SQL (as optional extension)

Different segments are delivered by different professors, but the quality was high across the board. All videos can be sped up, which is amazing. (Seriously, I cannot live without this any more. I tried to take Tableau’s own training programme and they were delivered *SO* slowly it was torture!)

I thought Udacity’s case examples struck a fair balance between the simplicity needed for pedagogical exposition, while preserving a reasonable amount of “real-world” relevance. I’m sure that my professional experience helped me to “fill in the gap” between these simple examples and the broader range of possible applications.

That said… I did get left with a feeling that a little *too* much had been glossed over.

I’m in the very fortunate position of having studied econometrics to Masters level. Don’t get me wrong – that was 10 years ago, so I am making NO presumptions about my expertise in this area! But I was pleasantly surprised to discover how much of that material I still had stored somewhere in my brain.

In the areas I remembered most about: linear regression and ARIMA time-series modeling, I definitely felt relevant nuance was omitted in the course… Important stuff that helps you assess whether you’ve chosen the right model for the job and how much you can lean on it. Students without that background may be left with an inflated sense of confidence about their abilities.

Does this mean you’re not taught anything serious? No.

I do think you are taught enough to be able to be functional in a junior role from day 1. But your development will depend on having more experienced talent to guide and supervise you from there on… or you doing more studying!

I should add that this issue is probably not unique to Udacity. I’ve seen other thought pieces observing that the speed and flexibility of these courses (and frankly, the need to keep clients happy so they give good reviews), creates incentives that push content a little more towards the lighter side.

Taking the level of detail as it was, I think the exercises gave me a fair chance to test and apply my skills. They are not totally easy, but I found the absence of serious challenges (even if optional) a little lacking. The feedback you get on your assignment submissions is good… but maybe with too much guidance.

If you have not given the exercise an honest shot or you mistakenly took an incorrect angle at the beginning, you basically get given the answer… But, on balance I think this is understandable given their monthly subscription business model and price point. (Of course, it’s also on you to be a responsible adult and study honestly!)

I felt reassured that I was able to download all the materials and back them up. There’s not a one-click solution to do this, but I was OK taking an extra minute to do so at the beginning of each unit. In any case, after “graduation” you get access to everything online for an entire year, so there is no rush.

Also, while engaging with the platform, I discovered that Udacity have this massive library of FREE materials… And I also discovered that many of these free materials had been repackaged into the course I was taking…! So, what was I paying for? The software licenses, and the opportunity to have work reviewed, access to mentorship, a Slack alumni network, and having a certificate issued at the end.

Now, I think that’s fair… but dedicated learners who are ready to make the extra effort to showcase their skills, really ought to look at the free materials first. They should see if the materials go far enough for them to develop a credible online portfolio without them paying for the certificate and work reviews.

And a bit about Alteryx...

The focus on Alteryx does need to be addressed.

I will write a full review of the product in due course… Suffice to say here, that Alteryx does an impressive job to lower barriers to entry around data-munging and descriptive and predictive analytics.

This is fantastic! And you’d think this would make it an excellent steppingstone on which to start your predictive analytics/machine learning journey… Except, it’s not!

Alteryx, sadly, is NOT a tool for the “citizen data scientist”, nor for the small-to-medium sized start-up. Why? Because it is EXPENSIVE!!! ($5,195 per user, per year, with no monthly option – to be precise!)

I judge that Alteryx is targeting large companies who have the scale to hire a large, specialised analyst base, where concerns about lowering the costs of onboarding and talent churn are material. I can see how they would be allayed by Alteryx, since it is easier to get junior talent started with it, and to be able to pick up other people’s work.

If you know you’re going to apply for such a company, this is all fine. But as you can tell from what I’ve said, they are not in the majority.

Putting cost aside, yes, Alteryx could be a great *steppingstone* into machine learning and analytics… but as your experience increases, I can’t help but think the cookie cutter approach to certain functions will eventually come to feel limiting. Yes, you can customise your own Alteryx tools in R (soon, Python as well)… but when I get to that point, I’d be asking myself: “Why am I using Alteryx?” I know the visual layer just slows everything down relative to coding-led alternatives.

For these reasons, I think the exclusive focus of Udacity’s course on Alteryx comes with bigger restrictions than I had initially appreciated.

No real regrets though. Alteryx gave me a reassuring introduction showing me that this stuff was not going to be out of my league. Also, I believe that learning different tools acquaints you with different grammars of thinking. This ultimately helps you consolidate your understanding in deeper ways. And as someone who frequently has to think about the UX side of tools I develop for my clients, I enjoyed doing some reflection on how well the product was put together to accomplish its goals.

Conclusions

Learning is iterative.

When you approach a new area of knowledge, understanding it helps you become more aware of what knowledge is more/less relevant to your goals.

I’m happy I took this course. But after realising all the limitations that came through the focus on Alteryx, I do think my time might have been better spent on another course instead (even though I was always committed to a multi-course learning journey). I could then have supplemented the other course with the free resources Udemy makes available, if needed.

Yes, it was nice to have this authoritative curation of Predictive Analytics tools… but now that I’ve introduced you to that distinction in this article, you can research it for yourself. You’ll then know which parts of more generalised machine learning courses to focus on, right??? ??

On the plus side, I have to be honest, it was great to feel that I really have the competence to take this stuff on, perhaps (with practice) to a level that’s above average in the market.

Anyway, the learning continues!

For the participants in current and future service economies, the old model of learning stuff in school or Uni, and expecting it to serve you your entire life, is gone! We’re just outrageously lucky that, at the same time, the internet and software is making it easier and easier for us to learn quickly, and acquire really cool, impressive, deliverable skills!

--------------------------

Roderick McKinley is a financial modeler, data-oriented business professional, and CFA Charterholder.

He works with a variety of clients to solve financial and operational business problems in diverse sectors ranging from FMCG, renewable energy, property, and even blockchain/crypto.

He blogs to contribute to the conversation about the emerging crossover between financial modeling and analytics. A trend he knows will unlock exciting developments in data-driven decision-making.

Roderick is also an experienced training provider offering bespoke programmes, and online courses. At time of publication, his courses on VBA, project finance, and renewable energy have been taken by over 12,000 students worldwide.

A travel enthusiast, Roderick is also a keen proponent of the remote working lifestyle, having based himself out of various locations in Asia, Europe, and the Americas since 2017.

You can reach out here on LinkedIn, or through his website: https://rmckinley.net