The Scientific Investor
Welcome to the Scientific Investor.
This blog should fill two purposes: It gives access to financial research recently published in the top academic journals and provides an intro to dig deeper into a specific topic (also in the form of a potential Bachelor, Master, and PHD thesis for students).
Moreover, it should inspire practitioners to use these academic insights for the implementation and refinement of investment strategies. It goes without saying that empirical results shown in the papers are based on historical data and there is no guarantee that a proposed strategy will be successful in the future.
never miss a new blog post. Subscribe to the Scientific Investor Newsletter.
THe SCientific Investor - PArt 8: Open Source CRoss-sectional Asset Pricing
A 2016 poll of 1,500 scientists conducted in the journal Nature reported that 70% of them had failed to reproduce at least one other scientist’s experiment (including 87% of chemists, 77% of biologists, 69% of physicists and engineers, and 67% of medical researchers).
Recently, also several voices in finance academia and practice have raised doubt on the truthfulness of empirical studies on the cross-section of average stock returns. As an example, Hou, Xue, and Zhang (2020) claim that more than 50% of characteristics – previously shown to be significantly related to average stock returns – fail the hurdle to display statistical significance at the 5% level in univariate portfolio sorts.
In a new research project, Chen and Zimmermann (2021) tackle the “replication crisis” in finance and advocate the idea of open-source cross-sectional asset pricing. They provide open-source code (in R and Stata) as well as data to replicate the relationship between 319 characteristics and the cross-section of average stock returns in the USA (during the original sample period and there-after).
The findings of the open-source research project are surprising and contradict the findings of previous meta-studies:
(a) For the 161 characteristics that were clearly significant in the original papers, 158 (98%) characteristics are shown to display an absolute t-statistic above 1.96 in univariate portfolio sorts (i.e., they are significant at least at the 5% level).
(b) For the 44 characteristics that had mixed evidence in the original papers, the reproductions find absolute t-statistics of 2.00 on average.
(c) A regression of reproduced t-statistics on original long-short portfolio t-statistics finds a slope of 0.88 and an R2 of 0.82. Hence, the replicated t-statistics match the original t-statistics based on economic significance.
(d) Abnormal returns due to the 161 characteristics decay after the initial sample periods of the original studies, but generally remain positive.
Hence, the paper takes a large step toward restoring trust in the literature of empirical asset pricing concluding that nearly 100% of the literature's predictability results can be reproduced. Being open-source, the research project also shows how these results are achieved.
THe SCientific Investor - PArt 7: Doctoral Education and Investment Performance
Pursuing a doctoral degree in Finance/Economic is a tough, but also rewarding route for ambitious students which consists of years of studying, researching, and teaching. Although usually designed for a career in an academic institution, a significant part of graduating PhDs make their way into asset management firms to manage other people’s money.
Are investment professionals with a PhD successful in doing so? And does their doctoral degree help them to outperform other investment managers (who do not have a graduate education)?
In their Management Science (2020) research paper, Chaudhuri, Ivkovic, Pollet, and Trzcinka empirically investigate the performance of investment products managed by firms in which PhDs play a key role (e.g., CEOs or CIOs) in comparison to products managed by otherwise similar firms. Their sample covers data from 59 quarterly snapshots from investment management firms in the period from 1993 to 2007; among others, they examine detailed performance statistics from 531 individuals holding a PhD degree.
The findings of the article can be summarized as follows:
(a) Financial performance of investment products managed by PhDs is significantly higher than performance of investment products by non-PhDs over the sample period. The difference in monthly (annualized) Carhart (1997) alphas amounts to 0.064% (0.768%).
(b) This positive performance spread gets larger if one considers the investment performance of products from firms which are founded by a PhD as the CEO or partner.
(c) Investment performance of PhDs with a Finance/Economics background is superior to performance of PhDs with a STEM (science, technology, engineering, and mathematics) background. Hence, field-specific training matters.
(d) PhDs who were successful in publishing in top scientific journals are also better investment professionals than PhDs whose scientific papers remain unpublished (i.e., unsuccessful PhDs).
(e) The performance difference between Finance/Economics professionals and STEM professionals is particularly large for unsuccessful PhDs; however, the difference is close for the top PhD students across disciplines. Thus, field-specific training is less important for the very top students across fields.
Do you want to dig deeper into the relationship between doctoral education and investment performance? Have a look at the working paper on SSRN or the published paper in Management Science.
THe SCientific Investor - PArt 6: how do venture capitalists make decisions?
Over the past 30 years, venture capital (VC) has been an important source for innovative companies. Firms that were supported by VC included Amazon, Apple, Facebook, Google, Netflix, and Starbucks – firms that had and have a tremendous impact on the US and global economy.
But how do venture capitalists (VCs) make their decisions? For their Journal of Financial Economics (2020) research paper, Gompers, Gornall, Kaplan, and Strebulaev surveyed 885 institutional venture capitalists (VCs) to learn how they form their choices among pre-investment screening, structuring investments, post-investing monitoring, and advising. The average VC firm in the survey is small, with 14 employees and five senior investment professionals.
The evaluation of the survey leads to the following important insights.
(a) The typical VC partner works 55h per week and spends 22h per week networking and sourcing deals as well as 18h per week working with portfolio companies.
(b) The average VC deal takes 83 days to close; the VC firm spends 118 hours on due diligence over that period and calls ten references.
(c) 15% of a VCs exits are through IPO, 53% are through M&A, and 32% are failures.
(a) A VC firm screens approximately 200 companies and makes only four investments in a given year.
(b) Over 31% of deals are generated through the VC’s professional networks and 28% of deals are pro-actively self-generated.
(c) When selecting investments, VCs place the greatest importance on the management / founding team, followed by the business model and the product. Company valuation is less important for early-stage deals, but becomes more important for later-stage deals.
(d) Only few VCs use discounted cash-flow or net present value techniques to evaluate their investments. The most commonly metric is the Multiple of Invested Capital (MoIC).
(e) 9% of all VCs and 17% of early-stage investors do not use any quantitative deal evaluation metrics. 44% of all VCs state that they also make “gut” investment decisions.
(f) The average expected IRR of a VC investment is 31% (!!!) p.a.
(a) After the initial investments, VCs provide a large number of services to portfolio companies, with strategic guidance, connecting investors, connecting customers, and operational guidance as the most important ones.
(b) When considering which of VC activities are most important for value creation, VCs rank deal selection as most important, closely before post-investment value-added and deal flow.
(c) When asked VCs what factors contributed most to their successes and failures of investments, the management / founding team is by far the most important factor.
Are you interested to dig deeper into the topic? Have a look at the working paper on SSRN (without charge) or the published paper in the Journal of Financial Economics.
THe SCientific Investor - PArt 5: Innovation and Product Differentiation of Mutual Funds
Believe it or not, there were more than 9,000 different mutual funds in the USA in 2020. Based on this large number, it is of first-order importance for funds to differentiate themselves from competitors. A potential way to position a fund as a “unique” investment vehicle is to apply a distinct wording in the fund prospectus.
As an example, the 2012 founded Patriot Fund was described as follows:
“The investment seeks growth of capital. The fund generally invests in common stocks included in the Standard and Poor's 500 Index, excluding those that fail the adviser's "patriotic" investment screen. The adviser's patriotic investment screen eliminates common stocks issued by companies doing business in nations supporting terrorism (“Terror Nations”), as defined by the U.S. State Department. Currently, there are several countries designated under these authorities: Cuba, Iran, Sudan and Syria.”
In their Journal of Finance (2020) research paper, Kostovetsky and Warner examine innovation and product differentiation using a uniqueness measure based on textual analysis of fund prospectuses. The textual uniqueness metric counts how many words a specific fund prospectus shares with other funds in the same Morningstar category. Consequently, funds that show a low word overlap are labelled as “textual unique”. As an example, the Patriot Fund is among the top 5% textual unique funds in the data sample.
What are the main findings of the study?
(a) Textual uniqueness identifies funds of certain sub-investment styles of the Morningstar category, such as, “thematic”, “socially responsible”, or “tax-aware” funds.
(b) Textual unique funds are particularly offered by small family firms: The start rate for families in the lowest size quintile is 168% higher than the start rate for families in the highest quintile.
(c) Textual unique funds are expensive: A one-standard deviation shock to text-based uniqueness is associated with a 12 basis points increase in the annual expense ratio.
(d) For new funds, textual uniqueness is related to higher money inflows: In the first 36 months, first with above-median textual uniqueness attract about 50% more in asset flows then comparable below-median textual uniqueness funds.
(e) The relationship between textual uniqueness and future performance is slightly negative; a one-standard increase text-based uniqueness is associated with a decrease in future average returns of approximately 10 basis points.
(f) Finally, textual unique funds have a lower flow-performance sensitivity, i.e., money inflows and outflows from investors are less dependent on the fund’s performance.
THe SCientific Investor - PArt 4: attention to global warming and Stock Returns
On December 28, 2017, former US president Donald Trump twittered:
“In the East, it could be the COLDEST New Year’s Eve on record. Perhaps, we could use a little bit of that good old Global Warming that our country, but not other countries, was going to pay TRILLIONS OF DOLLARS to protect against. Bundle up!»
In his quote, Trump confuses a short-term local weather phenomenon with global climate change. Indeed, this is a very common mistake made from humans around the world: One extrapolates personal experience about attention-grabbing weather events to long-lasting global trends. As a consequence, abnormally warm local temperatures often serve as “wake-up calls” that remind investors to the alert of global warming.
In their Review of Financial Studies (2020) research paper, Choi, Gao, and Jiang test how people react to abnormal local temperatures by examining their attention to climate change and stock prices. This is done using a worldwide sample from 2001 to 2017 based on 74 cities with major stock exchanges (among them New York City, HongKong, and Frankfurt).
The findings of the paper can be summarized in the following way:
(a) During abnormally warm months in a particular city, investor attention to climate change significantly increases. To measure investor attention to climate change, the authors use the volume of Google searches to the term “global warming”.
(b) During these abnormally warm months, local stocks of carbon-intensive firms underperform firms with low carbon emissions by approximately -0.4%. This result is consistent to the idea that if investors direct their beliefs towards global warming, they may buy stocks from low-emission firms and sell stocks from high-emission firms. They do so because they either want to ethically avoid holding social-irresponsible stocks or are concerned about lower future cash-flows through tighter environmental regulations for high-emission firms.
Which investor group is driving this effect? In line with the conjecture that individuals are particularly prone to limited attention and own personal experience, retail investors (and not institutional investors, such as pension funds) decrease their holdings of high-emission firms in an abnormally warm month the most. It is interesting to note that all these results only hold in a more recent time period and not in a “placebo time period” from 1983 to 2000, when global warming was less of an issue.
Have you become interested in the fascinating relationship between attention to global warming and stock returns? Read the full paper on SSRN (without charge) or the published paper in the Review of Financial Studies.
THe SCientific Investor - PArt 3: Empirical Asset Pricing via Machine Learning
Machine learning (ML) has emerged as a useful device in our everyday life, enabling the recognition of images and speech, performing medical diagnosis, and improving product recommendations. Is machine learning also useful for the design of investment strategies and can we use it to predict future returns of financial securities?
What is ML? The definition of this term is frequently dependent on the field. For research purposes in asset pricing, I define ML as a “collection of high-dimensional models for statistical prediction where (i) the risk of in-sample model overfitting is mitigated and (ii) efficient algorithms search among potential model specifications.”
In their Review of Financial Studies (2020) research paper, Gu, Kelly, and Xiu apply ML methods to predict future returns using a sample of 30,000 individual US stocks over a period of 60 years from 1957 to 2016. To predict future returns for each stock they use 94 characteristics, eight aggregate time-series variables, and 74 industry-sector dummy variables (totaling more than 900 prediction signals).
The following ML models are applied:
a) Baseline: Simple unconstrained linear regression
b) Penalized linear regression: Punishes the inclusion of new predictor variables and reduces potential overfitting of the model (models: lasso and ridge regression)
c) Dimension reduction techniques: Average the impact of all potential predictor variables to an aggregate predictor (principal component regression and partial least squares)
d) Penalized generalized linear models that allow for nonlinearities in the predictor variables
e) Boosted regression trees and random forests: Nonparametric models that allow for interactions between the predictor variables
f) Neural networks (deep learning): Nonparametric models that use activation functions and different layers to account for nonlinearities and interactions between predictor variables
The authors train these different models over the first 18 years in the sample, validate them in the consecutive 12 years, and use the remaining 30 years of data for an out-of-sample analysis based on monthly forecasts.
The results of the study are in favor of ML being helpful in predicting future stock returns. In particular, neural networks are able to substantially increase the accuracy of future return forecasts. Speaking in numbers, a value-weighted long-short portfolio that takes positions based on stock-level neural network forecasts earns an annualized out-of-sample Sharpe ratio of 1.35. This is twice the performance of a leading regression-based strategy from the literature.
What are the important predictor variables that boost accuracy of return forecasts? There is strong evidence that price trends (past momentum and reversal), liquidity (stock size and volume), as well as stock volatility are driving the results. Moreover, interactions between these variables and the aggregate book-to-market ratio seems to be important for the success of the documented neural network trading strategy.
Note that the investigation refrains from the impact of trading costs which are likely to be substantial (as the portfolio formation of the investment strategy requires a high turnover).
Are you interested to dig deeper into the topic? Have a look at the working paper on SSRN or the published paper in the Review of Financial Studies.
THe SCientific Investor - PArt 2: anomalies across the globe: once public, no longer existent?
Factor investing has emerged as a major trading style over the past decade. Investment companies cater increasing demand for factor strategies from institutional and individual investors through actively-managed funds and (quasi-) passive smart-beta ETFs.
What is factor investing? I define factor investing as “the selection of securities based on certain attributes that have produced risk-adjusted outperformance over a prolonged period of time." Such attributes are also called anomalies and are either due to the compensation for systematic risk or due to mispricing. Prominent anomalies with established factor premia can be constructed, among others, based on a security’s size, value, momentum, and reversal (I will come back to some individual anomalies in later episodes of this blog). As an example, a self-financing portfolio consisting of US stocks with high momentum (i.e., past annual performance) on the long side and US stocks with low momentum on the short side delivers a risk-adjusted return of 9% in the period from 1927 to 2019. The number of observed anomalies has grown tremendously over the past decade with a current count of more than 450 in the US stock market.
In this blog post, I focus on the question whether stock market anomalies are long-lasting or whether their profitability is destroyed after being made public by an academic research paper. I first turn to evidence for the US market. In their Journal of Finance (2016) paper, McLean and Pontiff document that – when comparing the initial in-sample period and the after-publication out-of-sample period – the magnitude of average factor premia decreases from approximately 7% to 3% per year, i.e., a decline of more than 50%. This finding provides evidence that investors learn from academic research which makes factor premium profitability disappear.
How is the situation in other stock markets worldwide? Jacobs and Müller (2020, Journal of Financial Economics) investigate the profitability and predictability of factor premia in 39 international stock markets. Their main findings are:
a) Most of the factor premia that are documented in the US are also observed on stock markets worldwide. Hence, investors around the globe have the tendency to (mis-)price stocks in a similar way.
b) Average factor premia of the initial in-sample period are slightly higher in the USA (7% per annum) than in the rest of the world (5% per annum).
c) The decay in factor premia in the after-publication period is significantly larger for the US than for most non-US countries. Hence, on average, factor premia tend to be more persistent in non-US stock markets. (This finding is consistent with the idea that learning about factor premia is particularly pronounced by sophisticated institutional investors that trade on US markets).
To summarize, academic research documents that academic research diminishes the magnitude of factor premia in the USA, but not so much for most other international countries. Consequently, a long-term factor investing strategy (based on known stock market anomalies) should be performed on a diversified global sample.
Note that in this article, we have refrained from two additional important practical considerations: (i) trading costs in the composition of the long-short factor portfolios (which can substantially decrease the net performance of factor premia) and (ii) short-selling restrictions of stocks (which can make the composition of the long-short factor portfolios impossible).
To dig deeper into the topic, retrieve the complete working papers on SSRN (McLean and Pontiff; Jacobs and Müller) and the published papers in the Journal of Finance and the JFE.
THe SCientific Investor - PArt 1: Cross-Asset Signals and Time Series Momentum
A lot of practical investment ideas are based on technical analysis, i.e., one uses past returns (and other characteristics) of an asset to predict its future return development. Let us extent this idea a bit: What about if we depend the decision to buy/sell an asset also on the past returns of another asset? In other words, can we use cross-asset signals to successfully derive a systematic trading rule?
In their Journal of Financial Economics (2020) paper, Pikäjärvi, Suominen, and Vaittinen investigate this idea based on a cross-asset strategy of the equity market (E) and the bond market (B). They show that (1) high bond market returns predict high future equity market returns and (2) low equity market returns predict high future bond market returns on a monthly horizon and beyond.
Why do we observe these relationships? The reason for this cross-asset momentum is mainly based on the notion of slow-moving capital and the perspective that equity is more risky than bonds. During general market upturns, investors shift their investment allocations from bonds to equity to receive a higher premium. During general market downturns, they divest from equity and invest into bonds to reduce potential future losses. However, these transactions do not occur immediately but are slow-moving and hence generate medium-term cross-asset momentum.
A simplified equity-bond cross-asset trading strategy at the beginning of month t can be constructed as follows: Compute the past 12-month equity return (E past) and the past 12-month bond return (B past). If:
a) E past is positive and B past is positive: Buy equity
b) E past is negative and B past is negative: Sell equity
c) E past is negative and B past is positive: Buy bonds
d) E past is positive and B past is negative: Sell bonds
e) Options a) to d) are not representative, invest in the risk-free rate.
Hold the portfolio for one month and then repeat the same procedure in month t+1.
Pikäjärvi, Suominen, and Vaittinen (2020) show that a diversified cross-asset strategy over 20 countries (including the USA, Germany, and Switzerland) from 1980 to 2016 outperforms pure equity or bond momentum strategies. It also has a significant higher Sharpe ratio than a buy-and-hold portfolio consisting of bonds and equities.
To dig deeper into the topic, retrieve the complete working paper (without charge) on SSRN or the published paper in the JFE.