Replicating Our Research in Europe

The gold standard of quantitative research is to test the same theory on out-of-sample data and find similar results over different time periods and in separate markets.

We have tested our core hypotheses over 48 years of data in the US, starting in 1964. We replicated our findings in a study of 23 years of data in Japan, starting in 1991. We not only published our findings, but we also built funds to trade on these strategies: funds whose performance is the ultimate out-of-sample test of our ideas.

Now, we have completed yet another replication of our findings, testing our hypotheses in European markets over 21 years of data, starting in 1997. We ran the same factor model and debt paydown algorithm that we had already tested in the US and Japan on this new European data with no adjustments. And we found significant evidence that our strategy of investing in small, cheap, leveraged companies that are paying down debt is robust across markets.

We are excited to share these results with you. We start with a reprise of our core hypotheses and then the results of our tests in Europe.

Our Hypotheses
The pioneers of the leveraged buyout industry generated spectacular returns in the 1980s, 1990s, and early 2000s, dramatically outperforming the public markets. We believe the core engine of those returns was neither prowess at due diligence nor operational savvy but rather the use of debt to finance the purchase of companies that were significantly smaller and cheaper than companies in the public equity markets.

We believed we could achieve similar success by applying those key quantitative criteria to public equity markets. Our hypothesis was that public equities that were significantly cheaper, substantially more leveraged, and much smaller than the broader indices would show the same outperformance over the broader market as these early LBOs. Our paper “Leveraged Small Value Equities” describes a simple stock ranking algorithm for selecting small, cheap, highly leveraged public equities and argues that investors could have outperformed private equity returns by using this strategy in public markets.

We believe that one of the core drivers of returns within private equity is the process of deleveraging. Private equity firms lever up the companies they buy and then use the cash flows of the company to pay down that debt. Because public equity investors can’t force public companies to pay down debt, public equity investors need to be able to effectively target the companies with the capacity and intention to deleverage. So we developed a machine learning algorithm that forecasts debt paydown among leveraged equities with up to 70% accuracy over a one-year horizon, which we describe in our paper “Forecasting Debt Paydown Among Leveraged Equities.”

Our findings in the US, described in those two key papers, were promising. Our out-of-sample tests in Japan provided evidence in support of our ranking algorithm’s ability to sort stocks by expected return and confirmed our debt paydown algorithm’s ability to forecast corporate deleveraging.
The next big test was European data, and we are delighted to report the results of this latest out-of-sample test.

European Results
Figure 1 presents an out-of-sample test of our US ranking algorithm, which sorts companies by characteristics such as value, leverage, size, and profitability. The results presented in Figure 1 suggest that this algorithm effectively ranks companies by their expected return when applied to new data. Between June 1997 and December 2017, annual portfolios of the top quartile of stocks in Europe would have outperformed the lowest ranked quartile of stocks by 4.9 percentage points per year. A diversified strategy of selecting the 50 highest ranked European stocks in each year would have earned an annualized return of 17.7%, outpacing the Fama/French Europe Small Value Index by 4.5 percentage points per year.

Figure 1: Out-of-Sample Test of US Ranking Algorithm in Europe, Jun 1997 – Dec 2017

Sources: S&P Capital IQ, Ken French Data Library, and Verdad Research. The universe of leveraged value stocks is formed from the cheapest 25% of the market and the most levered 50% of the market each year, excluding microcaps.

For context, the 4.5% premium of leveraged small value over small value is similar in magnitude to the 3.8% premium that we observed in our US research from 1965 to 2017. While the premium in the US includes in-sample data that was used to train the ranking algorithm, the observation of a 4.5% premium in Europe is completely out of sample.

Next, we tested our debt paydown algorithm on European data over the same period between June 30, 1997, and December 31, 2017. This algorithm was trained on 48 years of US data from 1964 to 2012 and achieved up to 70% accuracy in predicting debt paydown over a one-year horizon in the US. We also found a positive relationship between debt paydown probabilities and returns in the US.  

Figure 2 presents results of the out-of-sample test of our debt paydown algorithm in Europe. We sorted the universe of European leveraged value companies into deciles according to their probability of paying down debt over the next year, and we also show the top 5% and bottom 5% for reference.

Figure 2: Out-of-Sample Test of US Debt Paydown Algorithm in Europe, Jun 1997 – Dec 2017

Sources: S&P Capital IQ and Verdad Research.

As expected, the algorithm is effective at predicting debt paydown over a one-year horizon in Europe. There is also a positive relationship between debt paydown probabilities and returns in Europe. That is because companies that are most likely to pay down debt end up paying off a higher proportion of their debt over the next year. Similar to paying off a mortgage, companies that deleverage tend to experience an increase in equity value over time.

The results presented in Figure 1 and Figure 2 are fully out of sample, based on new data. No adjustments were made to either algorithm to account for local information in European markets. We believe these out-of-sample results provide significant evidence that our strategy of investing in small, cheap, leveraged companies that are paying down debt is robust across markets.

Could we potentially improve our quantitative process by incorporating the debt paydown probabilities into our ranking algorithm? Would a more parsimonious model that only considers the most robust signals in Europe provide better performance? We believe the answer to both questions is yes. We will present the results of that analysis next week, when we describe Verdad’s model evaluation process.