Lessons from our 2018 ‘Super Quant’ internships

Lessons from our 2018 ‘Super Quant’ internships

05-12-2018 | Visión

Top-notch investment strategies require top-notch research. The Robeco ‘Super Quant’ internship research projects, in areas such as language analysis or survey-based macroeconomic data, are part of this ongoing effort.

  • David Blitz
    Head of Quant Research
  • Martin Martens

Speed read:

  • Super Quant’ internships produce valuable research
  • In 2018, we looked at links between companies and spillover effects
  • We also analyzed SEC regulatory filings and macro surprise indices

In its effort to keep offering top-notch investment strategies, each year Robeco’s quantitative research department runs several research projects with our ‘Super Quant’ interns, under the supervision of our experienced researchers. Internships typically last for six months and are combined with writing a master’s thesis on the same research topic. For Robeco, these internships represent a unique opportunity to hire students from some of the best universities in finance and econometrics, and to either drill deeper in our existing intellectual property or explore new areas of research. This article describes the key findings of three of these internships in 2018.

Manténgase al día de las últimas perspectivas sobre inversión cuantitativa
Manténgase al día de las últimas perspectivas sobre inversión cuantitativa

Analyzing links between companies

This project looked into the links between companies or industries that are not fully incorporated in the prices of financial assets. The idea was to uncover and exploit a potential indirect momentum effect that spills over to the companies we can invest in. Such an indirect momentum effect could enhance our existing stock and corporate bond selection models.

To check whether an indirect momentum effect could provide valuable information, we used monthly data and regressed individual equity or industry returns on one-month lagged industry returns, using a Lasso (least absolute shrinkage and selection operator) regression method.

We made predictions both at industry level, leading to an industry rotation strategy we called ‘industry to industry’, and at individual equity level, leading to a strategy we called ‘industry to company’. The analysis showed that the ‘industry to company’ signal is stronger than the “industry to industry’ signal.

We analyzed 353,173 filings, which amounted to 20 million pages and 5 billion words

Browsing regulatory filings

This project investigated whether variables derived from the text in annual and quarterly reports may provide useful information for equity and credit investors. To this end, we downloaded all the 10-K and 10-Q files available from the EDGAR database of the US Securities and Exchange Commission.

After some adjustments – to remove numbers, symbols and punctuation marks, for example – we looked at several variables, such as text length, readability and sentiment. In total, we analyzed 353,173 filings, which amounted to 20 million pages and 5 billion words. The processing time was approximately 8 hours. To put this all into perspective, the average financial analyst reads 200 words per minute and would therefore need 50 years to digest all this information.

Our study showed that text analysis can be used to automate and speed up the reading process and that text variables are informative for a firm’s future equity and credit performance, mostly concerning volatility. In conclusion, automated text processing adds value compared with manual methods used in the past.

Economic momentum and macro news surprises

This project analyzed the use of macroeconomic data to predict equity, bond and currency returns. Most statistics are published with a lag and many academic studies argue that equity returns can predict GDP growth, but that the opposite is not true.

Recent academic work came to different conclusions: macroeconomic data may provide useful investment signals after all. Our study confirmed this, finding for instance that currencies from countries with the best economic momentum outperform those of countries with the worst economic momentum.

This project also looked at so-called ‘surprise’ indices produced by brokers. For many macroeconomic statistics, economists are polled ahead of publication. The ‘surprise’ is the difference between the predicted and the actual outcome. One feature of these indices is that they are not flat, but rise and fall over time. This implies that surveys go through overly pessimistic and overly optimistic periods, which, in turn, can be used to predict equity and bond returns.

Read more about our Super Quant internship program.

1See for example: M. Dahlquist and H. Hasseltoft, 2017, ‘Economic Momentum and Currency Returns’.