Over the years we have been evaluating and monitoring many alternative datasets. One of the most promising has been data from directors’ transactions. After revisiting the results from a project that we undertook a few years ago on this topic, we were encouraged by the consistency of results with an updated dataset, and its performance during the out-of-sample period. After finalising the design of the signal, we incorporated it into our investment process during the first half of 2020, as an additional measure in the Company Management stock selection component.
The Jupiter systematic Team

The Jupiter Systematic team, headed by Amadeo Alentorn, and including Tarun Inani, James Murray, Yuangao Liu, Sean Storey and Matus Mrazik (from left to right)

There is strong academic support for the usage of information based on directors’ transactions in their company’s stock, with one of the main arguments being information asymmetry. This argues that company directors enjoy information advantages over external investors about the firm’s profitability and prospects. The fact that they have disproportionally more information than outsiders can result in information about their transactions in their company’s stock, when noise is removed, being very powerful. The vast majority of academic literature focuses on how future stock returns are related to various metrics of such trades, director characteristics, firm level characteristics, etc. There are also ample empirical studies, and practitioners’ experience, evidencing future positive returns by following company directors’ transactions in their stocks. Their focus has tended to be on how different risk factors, director types and dynamics, confirmatory or contrarian trading of trades, etc., would be able to differentiate predictability of such transactions.

We have been considering the addition of directors’ transactions in our investment process for some time. The first exploration started in early 2016 and we have been monitoring the out-of-sample performance since. Our focus is mainly on whether such a signal would provide complementary information to our existing model, how to best extract information from each of those directors’ transactions, and whether such a signal would have sufficient in- and out-of-sample history for us to be confident an inclusion into our alpha model would benefit our investment process.

Directors’ transactions in their companies’ stocks can be triggered for various reasons. Needless to say, not all transactions contain an equal amount of information (some may have no information at all) that can be used to predict price performance of stocks in the future. However, it is reasonable to argue that such information may be related to company management teams’ decisions they make in their professional lives. Since those transactions are carried out on their own financial assets at their full disposal, they are voting with their own feet. Information extracted from these transactions may be expected to be additive to existing company management information extracted from financial statements, for example.

Many features of directors’ transactions can be important when determining the strength of the information of those trades. From a top level, obviously, sector, country, region and cultural differences could all have some impact; and, as a result, building in regional levels, with consideration for sector and country classifications, would be preferable.

Then, characteristics of each trade matter: for example, different transaction types – buy, sell or awards – would have quite different impact. Size of trade in both dollars and percentage of the director position being traded can also be important; very large trades (e.g. those above US$5m) should perhaps be treated differently.

Size of the trade compared to last 24 months of activity of the director is also a consideration. Also important is the trade flag such as tax related, automated sell or trades related to an employee compensation plan.

In addition to features of the trade itself, characteristics of the stock owner who initiated the trades, i.e. the director, are also critical. Those include seniority level within the firm (e.g. a CEO’s trade would have more information than a lower-level director); whether the insider has a finance related role (e.g. a CFO or finance director would perhaps contain more information than a director in a non-finance areas); how active is the director or group of directors (number of buys or sells of significance the director has made in the past two year); or whether they buy or sell on a subsequent price increase or decrease.

Last but not least is company characteristics. Many questions could be asked in this area, for example: does the valuation of the company trigger those transactions? Does size of the firm have a role in the transaction? What about beta, systematic risk of the firm? Was there a bull- or bear-run (both long-term and short-term) before the transaction? Is the company under any distress, or what is its probability of default?

All the above considerations have helped us in taking the seemingly random transactions made by company directors, and distilling them into a useful, proprietary stock selection signal. This new signal was introduced in the model in 2020 as part of the Company Management stock selection component. It has been designed to complement and diversify the existing signals within Company Management, which are mostly focusing on evaluating the historic corporate decisions made by a management team. With this new signal, we now also take into account decisions made by individual managers in their personal portfolio, as they trade in and out of the stock of their company.