Mispricing in leveraged value small-capitalization stocks

• Název práce v češtině: Chybné ocenění akcií s nízkou tržní kapitalizací a vysokým dluhem
• Název v anglickém jazyce: Mispricing in leveraged value small-capitalization stocks
• Klíčová slova: Anomálie, Prediktabilita výnosů, Testy oceňovacích modelů, Zadlužené společnosti, Hodnotové akcie
• Klíčová slova anglicky: Anomalies, Predictability of returns, Asset pricing tests, Leveraged equities, Value stocks
• Akademický rok vypsání: 2020/2021
• Typ práce: diplomová práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: Mgr. Martin Hronec
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 19.01.2021
• Datum zadání: 19.01.2021
• Datum a čas obhajoby: 15.06.2022 09:00
• Místo konání obhajoby: Opletalova - Opletalova 26, O314, Opletalova - místn. č. 314
• Datum odevzdání elektronické podoby: 02.05.2022
• Datum proběhlé obhajoby: 15.06.2022
• Oponenti: doc. Bc. Jiří Novák, M.Sc., Ph.D.

Seznam odborné literatury

Markowitz, H. (1952). Portfolio Selection. The Journal of Finance, 7(1), 77-91.

Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolios and Capital Budgets. The Review of Economics and Statistics, 47(1), 13-37.

Sharpe, W.F. (1964). Capital asset prices: a theory of market equilibrium under conditions of risk. The Journal of Finance, 19: 425-442.

Chingono B, Rasmussen D. (2015). Leveraged small value equities.

Chan, L.K.C., Hamao, Y. and Lakonishok, J. (1991). Fundamentals and Stock Returns in Japan. The Journal of Finance, 46: 1739-1764

Fama, E.F. and French, K.R. (1992). The Cross-Section of Expected Stock Returns. The Journal of Finance, 47: 427-465.

Chingono B, Rasmussen D.(2016). Forecasting debt paydown among leveraged equities.

Piotroski, J. (2000). Value Investing: The Use of Historical Financial Statement Information to Separate Winners from Losers. Journal of Accounting Research, 38, 1-41.

Bhandari, L.C. (1988). Debt/Equity Ratio and Expected Common Stock Returns: Empirical Evidence. The Journal of Finance, 43: 507-528.

Fama, E.F. and French K.R. (2015). A five-factor asset pricing model. Journal of Financial Economics, 116(1).

Banz, Rolf W. (1981). The relationship between return and market value of common stocks, Journal of Financial Economics, 9, issue 1, p. 3-18.

Eisfeldt, A. L., Kim, E., Papanikolaou, D. (2020). Intangible value. NBER Working Paper Series, (28056).

Mclean, R. D., Pontiff, J. (2016). Does Academic Research Destroy Stock Return Predictability? Journal of Finance, 71(1), 5–32.

Předběžná náplň práce v anglickém jazyce

Motivation

Asset prices should equal the sum of discounted future cash flows. Discounted, so when valuing an asset, the pitfall is not only to correctly predict the future cash flows but to determine the exact discount rate such that investors level of risk aversion and opportunity cost are really reflected. Capital asset pricing model (CAPM) by Sharpe (1964) and Lintner (1965) had served for this purpose for decades. However nowadays, there is substantial amount of evidence on various factors that capture more variance than the single variable - market proxy of CAPM. Examples of the most prominent factors affecting returns on top of the CAPM framework include size effect noted by Banz (1981), leverage effect identified by Bhandari (1988) or value effect uncovered by Chan, Hamao & Lakonishok (1991) and many other anomalies. As a result of the ability of those factors to further explain returns, multi-factor models taking the implied premiums into account are preferred in asset pricing literature. The mostly referred to are three-factor or five-factor models by Fama & French (1992) and (2015) respectively. In spite of the general acceptation of these models, plenty
of anomalies is still left behind and non-zero α-returns can be observed even in these models if the right factors and strategies are applied, e.g. reflecting intangibles along with fundamental analysis yields positive α even in five-factor model with augmented momentum factor (Eisfeldt et al., 2020). On the other hand, pool of academic literature focused on detecting anomalies and successful trading strategies has became very saturated and more importantly the historically discovered patterns are getting reflected in the asset prices (Mclean Pontiff, 2016). Example of such factor whose premium vanished over the course of last decade is the "value effect".

I would like to take a closer look at companies exposed to some of the historically most impactful factors mentioned above - value, size and leverage, and investigate whether 1) such universe still "enjoys" the premiums it used to and more importantly 2) there are at least some "unexplored betas" within the universe, e.g. if we are able to consistently pick firms with excess risk-adjusted returns based on various factors, there must be a variation not captured by standard asset pricing models such as five-factor model etc. The latter is the focal point of this thesis as there is a lot of space for potential mispricing by investors in this universe.

Since Piotroski (2000) argues that success of the value strategy, portfolio with high book-to-market ratio in this case, originates from the high return of only a few companies while majority does not even produce positive risk-adjusted returns, some of the value companies are probably heavily mispriced. In case of leveraged firms, it is their ability to repay their debt that separates the winners from losers (Chingono & Rasmussen, 2016) and this ability might be subject to incorrect assessment by investors, especially with respect to different credit cycles. In addition, firms with smaller market capitalization are subject to scarce coverage by sell-side analysts increasing the probability of mispricing by retail investors by incorrect prediction of future cash flows.

Leveraged firms with favourable valuation multiples and small market capitalization appear to create the pool of stocks 1) whose discount rate truly reflecting investors preferences might by subject to yet undocumented factors and 2) where incorrect evaluation of future cash flows by investors might occur. Therefore, I would like to construct stock ranking system within such pool of stocks with emphasis on ability to repay debt and fundamental analysis to show if this is the right environment to implement mispricing strategy.

Hypotheses

Hypothesis #1: It is possible to predict firm’s ability to pay down debt in the
universe of leveraged small-capitalization firms.
Hypothesis #2: Investing strategy based on the constructed stocks ranking
mechanism yields excess risk-adjusted returns.
Hypothesis #3: Mispricing by investors regarding the leveraged small-cap high
value firms differs with credit cycles. E.g. the constructed stocks picking
system performs better during periods of high credit spread.

Methodology

Yearly or quarterly international cross-section data from the most recent decades from selected markets will be used for the purpose of this study. Both company level and macroeconomic data will be extracted from Thomson Reuters Datastream.

First step in the analysis is to select the criteria for the universe of leveraged value small-caps. Even though small-capitalization firms are usually defined as those with total equity value between USD 300 milion and USD 2 billion, the definition in this study might slightly differ such that the resulting number of firms in the constructed universe is appropriate for the analysis. Subsequently, the companies will be ordered based on a debt-related ratio and a valuation multiple in order to determine those leveraged and containing value, most likely based on a percentile boundary. As a result of the intersection of the small, value and leveraged sets of companies, universe of stocks for further analysis will be created.

Since debt reduction is the most important factor in the process detecting mispriced companies within the leveraged value small-caps (Chingono & Rasmussen, 2015), separate model for prediction of the probability of debt repayment is to be constructed. Panel data linear regression and machine learning techniques will be applied to company financial data and forward-looking estimates of analysts. The following equation will be estimated:
yi,t = f(xi,t−1,1, xi,t−1,2, .., xi,t−1,K) + ϵi,t (1)
where y is dummy variable equal to 1 if the company i reduces its long-term debt in period t compared to previous period and 0 otherwise. xk denotes individual explanatory lagged variables, e.g. gross margin, growth rate of sales, asset turnover etc.

As the intended stocks ranking mechanism aims to predict and not to explain, again machine-learning techniques such as random forests and gradient boosting machines will be applied to predict returns in the subsequent period. It is defined as follows:
ri,t =g(yi,t, ct−1, x′i,t−1,1, x′i,t−1,2, .., x′i,t−1,K) + ui,t (2)
where ri,t is the return on stock i during period t, ct−1 is a variable capturing the efficiency of credit market in period t-1, yi,t is the estimated probability from equation (1) and x′ k denotes individual explanatory lagged variables.

Such and estimation procedure will be applied on T periods representing the in-sample part of dataset.

In the final stage, portfolio consisting of certain amount of stocks with the highest predicted returns based on equation (2) will be constructed and rebalanced each year in the out-of sample part of dataset.

Performance testing of the constructed ranking system will be based on CAPM and multifactor models by Fama & French (1992,2005) using the entire out-of sample period.

Expected Contribution

Historically we have been flooded with factors explaining the cross-sectional variation in returns. However, in many cases the findings can be attributed to specific datasets or the implied premiums just vanished through the following years as investors risk aversion and opportunity cost are functions and not constants. For example, the near-zero interest rate environment we have observed in recent years might have forced them to explore other investing options. Therefore, I would like to contribute to the existing asset pricing research with machine-learning stocks ranking system based on prediction of discount rates in such universe of stocks that have historically been positioned to yield premiums but now only rarely are found. Especially, the introduction of credit cycle explanatory variable in the prediction might unveil some variation in returns of value and leveraged equities. Moreover, I would like to reflect such relationships that might be only universe specific and identify companies that are most suitable for mispricing strategy as investors tend to incorrectly predict their future performance.

Analogous study focused on similar universe of stocks is Chingono & Rasmussen (2015, 2016). However, substantial differences and extensions should be noted. Especially, our machine-learning approach for construction of the stocks ranking system is expected to perform better out-of-sample due to the predictive power of this technique. Moreover, this study extends the geographical coverage as we are going to use international data so differences in predictability and repayment risks between individual markets could be observed. There will be also update and extension in terms of data recency so more up to date relationships could be taken into account during estimation and out-of sample testing. Last but not least, introduction of credit cycles in the models is expected offer better insight to investors risk aversion variability in time resulting in more accurate discount rates.

Outline
1. Introduction
2. Asset pricing theory and anomalies review
3. Data & Universe construction
4. Methodology
5. Results
6. Conclusion