How to Create a Risk-On vs Risk-Off Stock Market Dashboard in Python — Part III: Update | by Cordell L. Tanny, CFA, FRM, FDP | Jan, 2024

Cordell Tanny has over 23 years of experience in financial suppliers, specializing in quantitative finance. Cordell has beforehand labored as a quantitative analyst and portfolio supervisor for a primary Canadian institution, overseeing a $2 billion multi-asset retail funding program.

Cordell is at current the President and co-founder of Sample Prophets, a quantitative finance and AI choices company. He’s moreover the Managing Director of DigitalHub Insights, an instructional helpful useful resource dedicated to introducing AI into funding administration.

Cordell acquired his B. Sc. in Biology from McGill School. He’s a CFA Charterholder, a licensed Financial Hazard Supervisor, and holds the Financial Data Expert structure.

Go to trendprophets.com to be taught additional.

Hyperlinks to earlier elements on this sequence:

How to Create a Risk-On vs. Risk-Off Stock Market Dashboard in Python — Part I

How to Create a Risk-On vs Risk-Off Stock Market Dashboard in Python — Part II: Correlation Analysis

Good day all people! I do know a lot of you’re prepared for my subsequent instalment on this evaluation enterprise. I’m now pursuing one of many arduous phases: attribute engineering. It’s taking time as a result of quite a few steps involved. Nevertheless concern not! It’s coming.

The intention of this textual content is to supply an substitute and share what I’m engaged on. As a reminder, the target of this enterprise is to:

1) Share my evaluation course of and logic when working by way of a enterprise like this. It consists of:

a. Thought expertise.

b. Making a logical improvement and transfer diagram (it’s going to vary a variety of cases).

c. Decide wished sources, data, and technical requirements (programming languages, database setup, varia).

2) Share my information and experience of working as an skilled money supervisor and quantitative finance practitioner.

Up to now, we’ve received tried to know the connection between our risk-on and risk-off indicators. We’re using ETFs for various sectors and themes that must behave otherwise primarily based totally on when the market is in a greater hazard regime or a lower hazard regime. Please see the sooner articles for additional information.

So, what’s the next step? As beforehand talked about, I’m not a fan of correlations for a variety of causes. I nonetheless check out them because of they’re important, nevertheless they cannot current the whole description of the relationships in one of the best ways now we have to assemble an outstanding monitoring dashboard and hopefully an outstanding predictive model. Which signifies that now we have to ascertain completely different methods of assessing how these investments relate to at the very least one one different.

The following step is to assemble a hierarchical clustering algorithm to help visualize the relationships. This can even help us filter out the ETFs that really don’t help us the least bit. Nevertheless, to make a hierarchical clustering algorithm work we would like many choices. And we cannot merely use returns or correlations.

So proper right here is the place the idea expertise stage begins. What would you do? It’s good to generate on the very least 10–20 new choices to make a hierarchical clustering model a worthwhile practice. Everyone knows that we wish 2 clusters, nevertheless that doesn’t indicate we could have two clusters. So k-means is out since we’ve received to specify beforehand what variety of clusters we’re looking for. We are going to specify 2–3 clusters, nevertheless I don’t must introduce any bias into the model.

And at this stage, we solely have returns to work with. Thus, all our choices should be derived from returns. We are going to use the correlations as one attribute. Nevertheless proper right here is essential consideration, and the necessary factor message on this substitute. We’ve two factors to maintain:

1) We cannot use single time restrict estimates for our choices. And clustering algorithms ought to have one price only for each attribute and each ETF. So, we wish to have the flexibility to grab the time-varying nature of every attribute for each ETF!

2) The algorithm will attempt to search out the hole between every attribute of each ETF. How do you summarize the time-varying nature of each attribute in a single measurement?

That’s what I’ll downside the reader with. See should you occur to can provide the solution to maintain this. You probably may even submit throughout the suggestions proper right here as I work on the next half. I’ve good choices which I’ve used before now, nevertheless I can’t spoil the shock.

We’re once more to attribute engineering. I’ve to engineer additional choices that will fulfill the two components above using returns, and I’d want to hunt out completely different choices that aren’t primarily based totally on returns and that may even fulfill the two components above. That’s the pleasing half! That’s the place I sit with graph paper and a mechanical pencil and assemble ideas maps. They help me visualize what I’m contemplating.

That’s moreover a unbelievable exercise for ChatGPT. I’ll have conversations with it, being very thorough and cautious with my prompts. I give very explicit parameters for what I’m attempting to achieve, the foundations that should be glad, and it should be life like and doable to implement. I moreover don’t must bias the outcomes with what I must try. And this truly is the easiest half because of I like after I see ideas and methods I’ve under no circumstances heard of. That’s how I be taught and get greater.

That’s what I’m engaged on now. I’m researching a variety of methods to interrupt down our data into choices. The following article will present the outcomes.

Thanks for learning! And take note, should you want to share some ideas and enter into this enterprise, please accomplish that throughout the suggestions.