Today you have a completely different article in front of you than usual. Today we want to show you how we can identify companies that might need Falcon. This is not about marketing. It's about identifying companies with a specific need. The need we have picked out today is at the same time little known but all the more exciting: Shareholder Activism.
A shareholder activist is a person who attempts to use his or her rights as a shareholder of a listed company to bring about change within or for the company. The term refers in particular to investors who believe that the management of a company is doing a bad job. This group of activist investors often seeks to gain control of the company and replace management or force major corporate change. It is not uncommon for the well-being of the activist to take precedence over that of the company.
It seems that most German managers are not really familiar with the term. However, more and more aggressive minority shareholders are pushing for a sale or initiating hostile takeovers by competitors. The consulting profession has long recognized this. Strategy consultancies such as McKinsey or BCG, but also the Big Four and financial service providers such as PWC and A&M are intensively involved in defending against shareholder activities. They offer projects that can naturally be very well supported by Falcon. These are real transformation projects that cover the entire company.
Reason enough for us to start an attempt to identify companies that might fall victim to activists in the future.
How we achieve 90% prediction accuracy in 3 steps
Nobody knows exactly what fosters shareholder activism. The research identifies a whole range of potential influencing factors and disagrees about the direction and strength of their impact. Here you can read about it quite well.
The multitude of mutually influencing factors literally screams for modern methods of machine learning and that is exactly what we have done.
Step 1: Data
In a small field trial, we have put together a set of companies that, purely by size and industry, might fall into the clutches of activists. The number of companies is in the mid three-digit range. Nearly 29% of the companies have already been victims of activists. So they were already the target. Our goal was to develop a method that "understands" why a company was a target or non-target in a three-year period. If this method is good enough, we could test it on new companies and determine at the push of a button who might be a target tomorrow. Using publicly available sources, the next step was to create a data set which, in addition to the classic questions of size and industry, also contains various other variables that provide us with information on the net assets, financial position and results of operations of the companies.
A small - but representative - extract from the data shows that at first glance there is not really much difference between targets and non-targets. But that can be misleading. Because as beautiful as a 2D pair plot may look...the real connections certainly go beyond the second dimension. Nevertheless, it's worth looking around a little in the second dimension.
This is shown, for example, by this small correlation plot. There seems to be a positive correlation between targets and e.g. the average EBITDA margin. Alone this does not tell us anything. But it is extremely important to develop a feeling for the data. Highly correlated data in particular can cause problems in analysis. By their very nature, corporate KPIs are strongly correlated.
Step 2: Build a model
A statistical approach to establish the impact relationship between the target (yes/no) and the business metrics would be a logistic regression. Although logistic regressions are widespread, they have some serious drawbacks. Where potential characteristics are unknown, simply adding all characteristics is expected to reduce model performance due to a loss of efficiency. In addition, the model cannot capture a complex interaction structure.
Instead, we have opted for approaches that better master high-dimensional feature spaces. They usually lead to highly scalable and robust models. The following results show what we have used.
Schritt 3: Ergebnis checken
With our models, we have classified up to 90.75% of companies according to reality. This means that our best model correctly understood whether it was a target or not for almost 9 out of 10 companies using statistical methods. For the interested...below a few ROC curves.
But we can do a lot more. Not only can our model classify companies in just under 0.2 milliseconds, it also tells us which company characteristics are important for the decision. We don't want to reveal everything here. But then we reveal a bit: It's not just about being particularly good in terms of sales and earnings. In fact, companies seem to be of interest to activists who move to extremes - especially good or especially bad. However, some things always seem to favour activism: a shaky financing structure, high overheads (as well as SG&A) and a lack of innovation. The latter applies in particular to companies that have the necessary cash reserves but do not reinvigorate. We'll tell you a little bit more about our findings here.
Our model is up. We can expand the data base at will and identify companies that will have problems with their shareholders in case of doubt. And that's exactly how we look for companies that might need Falcon. For example, reducing overheads is wonderful with Falcon.