A Sharper Signal Amid the Noise, Part II: Evidence from Europe

Paying down debt is an important return driver across geographies and across time.
 

By: Brian Chingono

Last week, we showed that deleveraging is both important for returns and predictable to a meaningful degree, making it a ripe field for forecast improvements. We also demonstrated that improvements in the design of a deleveraging model can lead to higher out-of-sample accuracy by around four percentage points in the US.

Our findings in the US were based on 50 years of out-of-sample data. But the gold standard for empirical research is to find similar evidence in different markets, in addition to over long horizons. Therefore, we extended our research to Europe to evaluate the model’s accuracy on a brand-new dataset from across the pond.

The results of our out-of-sample test in Europe are shown below. The original model we shared in 2019 had 66% accuracy in predicting deleveraging over a one-year horizon among top-decile stocks. The updated model presented below has 69% accuracy in Europe, representing a three-percentage-point improvement.

Figure 1: Out-of-Sample Deleveraging Results in Europe (1997–2017)

Source: S&P Capital IQ and Verdad research. Eligible universe excludes net-cash firms.

Similar to the results we shared last week from the US, we find that companies with a higher propensity to pay down debt also reduce their debt balance by the greatest magnitude. And even though the model was not trained to forecast returns, companies in higher prediction deciles for deleveraging also tend to have higher forward returns. We believe the characteristics in the three right-hand columns help to explain this return pattern. Companies that are more likely to pay down debt tend to be cheaper, as shown in the Price/Book column. They also tend to have a moderate amount of debt on their balance sheet (2.5x Net Debt/EBITDA in the top decile), and this leverage amplifies the value premium over long horizons. In addition, deleveraging mechanically increases the equity balance over time, and effectively functions as a return of cash to investors.

Next, we wanted to see how the deleveraging probabilities from this algorithm would fit within a ranking model that’s designed to forecast returns. Ideally, the variables in a model should be uncorrelated so that each signal brings new information to the table when ranking stocks by expected return. A simple way to test for this independence is to measure the correlations between signals in a model, with correlations closer to zero implying that a pair of variables would complement each other in a regression model. In the figure below, we measured the correlations of our deleveraging probabilities against standard measures of value, profitability, and momentum–factors that are commonly associated with higher expected returns.

As shown in the results below, deleveraging probabilities are largely independent of two measures of value: EBITDA/EV and free cash flow yield. We believe this is good news because those two signals tend to be the workhorses of value strategies. So these results suggest that adding deleveraging probabilities to a simple screen that sorts companies by EBITDA/EV and free cash flow yield would provide uncorrelated information toward selecting potential winners.

Figure 2: Factor Correlations vs Deleveraging Probability, US and Europe
(1997–2017)

Source: S&P Capital IQ and Verdad research.

The deleveraging probabilities appear to have moderate correlations with Price/Book and gross profitability, whereby companies that are more likely to pay down debt tend to have lower Price/Book valuations and they tend to have higher gross profitability. The associations are not strong enough to imply that the information is redundant, but they do suggest that practitioners should account for this partial overlap in their weighting system if deleveraging probabilities are combined with Price/Book and gross profitability in a ranking model.

Having seen accuracy improvements of three to four percentage points when forecasting deleveraging in the US and Europe, we sought to measure how these forecasting improvements relate to expected returns. In the figure below, we start with a ranking model that sorts companies by deleveraging probability, value, profitability, and momentum. We then measure how annualized returns at the top of the screen change when we use the updated deleveraging probabilities versus the original versions of those deleveraging forecasts. Based on 20 years of data from 1997 to 2017, it appears that forecasting deleveraging more accurately by three to four percentage points translates to a 1.9% improvement in annualized returns, on average, at the top of the screen. For context, a strategy that targets cheap, profitable, levered stocks that are paying down debt would have returned 17% annualized in Europe since 1997, compared to 13% annualized in the Fama-French Europe Small Value Index. If the same strategy utilized the improved deleveraging forecasts, the evidence below suggests that the total return would be in the range of 18-19% annualized.

Figure 3: Excess Return from Improved Deleveraging Accuracy, US and Europe (1997–2017)

Source: S&P Capital IQ and Verdad research. Universe includes ~700 levered, value stocks each year.

We’re pleased to see confirmation of an economically logical and meaningful relationship between higher deleveraging and higher expected returns. While markets can swing based on unpredictable turns in the news cycle, it’s reassuring to see evidence pointing to the long-term merits of holding companies that are consistently paying down debt over the long haul.