Having a spouse in Web Analytics while I work in the Financial Software field has shown me several similarities between our two areas of work. In a nutshell we each:
- Work with time-series data
- Use statistics to find and explain patterns
- Work hard to turn complex data into useful, understandable, and actionable information
The time-series data in web analytics involves data such as visits per day, new versus returning visitors, average time on a page. Time-series data for financial applications involve stock, bond, and ETF prices, dividend events, and a bunch of other data like earnings announcements, government economic data, merger and acquisition data, etc.
There is so much financial data, that a financial analyst much choose to focus on a limited niche such as specializing on a geographical region (say Japan), or focusing on a specific industry (say Energy).
Luckily I am not a financial analyst, however I am involved in financial analytics. The difference? I write software that is designed to look at large sets of quantitative financial data in order to design more resilient investment portfolios. One of the primary data sets my software uses is total-return price data. [There are lots of other data the the software can use: P/E ratios, yield, beta to name a few. However I will focus on price data for the sake of brevity.]
Statistics and Pattern Recognition
In web analytics, finding patterns is important to making better web design and marketing decisions. In financial analytics, finding patterns is helpful in building stronger investment portfolios. One of the most useful patterns sought in finance is low-correlation asset pairs or asset groups. Correlation is a statistic that tells how the price-movements in one asset relate to the price movements in another. If two assets move together in exactly the same way they will have a correlation of 1.0, the highest possible. If they move independently of each other they have no correlation and thus a correlation of 0. If they move completely opposite each other they have a negative correlation of -1.0, the lowest possible correlation.
A common made-up example used to show how negative correlations can build a stronger, more diversified investment portfolio is the example of two stores on a small island called Markowitz Island. Store Sundown sells only sun tan lotion while Rainmakers sells only umbrellas. On Markowitz Island, about half of the days are rainy and the other half are sunny. On average both Sundown and Rainmakers earn 10% profit per year. However, on any given day each won will make money and the other will lose money. The profit/loss correlation between Sundown and Rainmakers is nearly -1. In this example, a portfolio invested half in Sundown and half in Rainmakers would produce a very steady stream of profits.
The other common statistic used in financial analytics is variance. The daily profit variance of both Rainmakers and Sundown would individually be rather high. However the profit variance of a 50/50 portfolio in each would be very low. Variance can come under various names: volatility, standard deviation, σ. These are all mathematically related to each other. The take-away is that lower is better when it comes to variance. Lower variance is associated with lower risk.
The “trick” in web analytics is to find associations (patterns) between actions that produce high ROI. The “trick” in financial analytics is to find associations between assets that zig when the others assets zag. There are a lot of mathematics behind finding optimal patterns, but those details are not important for this post.
Data is just data, until it becomes actionable. I see, with help from my portfolio analytics software, how pieces of a portfolio fit together like puzzle pieces. Math plays a big role in my understanding, and in the software I create. However, very few of my clients speak the language of math in which I am fluent. If they were they could create similar software and have little need of mine.
It is relatively easy to tell my clients what to do… buy this, sell that. This is seldom sufficient. My clients want to know why. I have a fluent and beautiful language to explain… math… but it is as a foreign language to most of my clients. It is not high school math, and it is not typical undergrad math. It is fairly advanced and specialized.
I thus look for a common language to communicate the whys. I have found a few methods that work — color maps, heat maps. Seven colors that roughly explain the purpose for each asset in a portfolio. Also correlation matrices — sophisticated clients understand these. Finally, back tests which show how effectively optimization data can predict subsequent results.
I also use the same portfolio optimizer to manage my personal investments. Together these factors fulfill several client needs:
- They increase Client (capital “C” Clients) confidence in the analysis and optimization results, allowing them to use the data to make investment decisions.
- My Clients can share portions of this analysis with their clients (lowercase “c”), showing them a thorough analysis of the “whys” of their investment decisions and increasing client confidence.
- Overall, assuming my portfolio optimization software provides benefit over the long term, both the Clients and their clients are happier, resulting in more business and more profit.
It is not sufficient to simply provide good recommendations. The recommendations must be explained in a manner that inspires confidence and results in the choice to implement some, most, or all of them.