Survey of Quality Investing Wednesday, 21 June, 2017

Interesting academic paper which analyzes different definitions of "Quality" factor:

Authors: Hsu, Kalesnik, Kose

Title: Survey of Quality Investing

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2971185

Abstract:

Factor investing has experienced a resurgence in popularity under the moniker “smart beta.” Several traditional factors, such as value, size, momentum, and low beta, are well defined and have been heavily researched in academia as return anomalies for many decades. These factors have also been exploited by practitioners as quantitative strategies for enhancing returns. Today, these factors each define a distinct smart beta category (think of style boxes for smart beta strategies) and are the foundational building blocks for the now-ubiquitous multi-factor products.

Notable quotations from the academic research paper:

"The recently popularized quality factor, however, appears to stand-alone in many regards. Like the four previously named factors (value, size, momentum, and low beta), quality investing has been widely practiced as an investment strategy by portfolio managers. MSCI, FTSE Russell, S&P, EDHEC, and Deutsche Bank, among others, have created quality factor indices for licensing and have generally included quality as a part of their multi-factor offerings. But, unlike the conventional factors, quality as a source of return has attracted limited academic attention and has been focused on only some facets of what practitioners categorize as quality. In a way, quality is a product waiting for academic validation, and the early results appear to be more inconclusive than its massive popularity might warrant.

In a routine product conversation with investors, the quality factor is pitched by providers as an independent source of return and as providing diversification due to its supposedly low correlation with the value factor. What remains uncomfortable for researchers, however, is that the quality factor is constructed very differently than other factors. Factors, such as value or low beta, are created from a particular stock characteristic (or a set of highly related stock characteristics) to capture a risk premium associated with an undiversifiable economic risk or to capture an anomalous return associated with a persistent investor behavioral bias. For example, the value factor is generally constructed from stocks that have high book-to-price, high earnings-to-price, high dividend-to-price, or some combination of these valuation measures. Regardless of the chosen definition for factor construction, the resulting portfolio looks and feels like a value portfolio in that it owns low valuation stocks.

In contrast, quality factor portfolios, as constructed by the different providers, have been entirely multi-signal in nature. Providers tag a stock as high quality if it scores high on some combination of the following attributes: earnings growth, earnings-growth stability, low return-volatility, high profitability, high return on assets (ROA), low debt ratio, and low accounting accruals. We begin our study by examining definitions of quality implemented in different product offerings. We show that quality, as executed by practitioners, is a collection of heterogeneous signals having little correlation with each other. Quality, as currently defined, would seem to be a catch-all bucket for those portfolios that blend many otherwise independent return factors.

These stock screens appear to favor heterogeneous groups of stocks and produce portfolios with low correlation to one another. The stocks appear to be selected for their diversity, and the multiple signals used in constructing the quality portfolios do not appear to be proxies for a single specific risk exposure or behavioral anomaly. Thus, a quality portfolio can seem more like a quantitative strategy based on multiple signals than a factor in the heritage of the arbitrage pricing theory (APT) framework. This unique feature plays an important role in how we analyze quality products versus how we examine other more conventional factors.

Because quality is being defined as it is—a collection of heterogeneous signals—a spectrum of possible portfolio outcomes exists. In the most positive case, the resulting quality portfolio has impressive out-of-sample performance if each of the signals or characteristics included in the construction of the portfolio represents a unique source of premium, whether risk based or behavioral based. In this case the resulting quality portfolio would be a multi-factor portfolio offering a diversified basket of excess returns.

In the worst case, the multi-signal portfolio will have indistinguishable from zero out-of-sample performance despite its impressive back-tested t-stats. Why could t-stats be misleading in this case? The large pool of multiple signals to select from creates opportunities for intentional or unintentional data mining overstating the t-stats; this issue has been emphasized by Harvey, Liu, and Zhu (2015). Further, from Novy-Marx (2016) we know that combining several uncorrelated factors selected because of their spurious in-sample performance further overstates the t-stat. For example, if an ex ante random strategy has an in-sample t-stat larger than 2 with only a meager 5% probability, then a mix of three such strategies selected ex post for best performance out of 20 strategy realizations will register a t-stat above 2 with a probability of nearly 98%. In reality, this portfolio will offer investors nothing more than noise and unwarranted fees and expenses, suggesting that the process of analyzing and contrasting different quality portfolio methods is both more difficult and more important.

Given the observation that quality is a collection of heterogeneous signals, we examine where the current quality portfolios are on the spectrum of robustness: from being a collection of robust anomalies with a high chance of outperformance, at one end, to being a collection of signals selected due to their spurious in-sample performance with little chance of outperformance out of sample, at the other end. We use the Hsu, Kalesnik, and Viswanathan (2015) method to identify robustness of variables used in quality definitions."


Are you looking for more strategies to read about? Check http://quantpedia.com/Screener

Do you want to see performance of trading systems we described? Check http://quantpedia.com/Chart/Performance

Do you want to know more about us? Check http://quantpedia.com/Home/About

Portfolio Weighting Schemes for Commodity Futures Risk Premia Wednesday, 14 June, 2017

Related to commodity trading strategies, mainly to:

#21 - Momentum Effect in Commodities
#22 - Term Structure Effect in Commodities

Authors: Rad, Yew Low, Miffre, Faff

Title: How Do Portfolio Weighting Schemes Affect Commodity Futures Risk Premia?

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2977710

Abstract:

We examine whether and to what extent successful equities investment strategies are transferrable to the commodities futures market. We investigate a total of 7 investment strategies that involve optimization and mean-variance timing techniques. To account for the unique characteristics of the commodity futures market, we propose a novel method of classification based on momentum or term structure properties in the formation of long-short portfolios in conjunction with the quantitative strategies from the equities literature. Our strategies generate significant excess returns and risk-adjusted performances as measured by the Sharpe and Sortino ratios and the maximum drawdown. We find no significant correlation between the strategies’ excess returns and common risk factors. There is no evidence that excess returns are a compensation for liquidity risk. The strategies are robust to transaction costs and choice of model parameters and exhibit stable performance across various market environments including times of financial crises.

Notable quotations from the academic research paper:

"There are theoretical and empirical reasons to believe that commodity futures investments command positive risk premia. The theoretical considerations relate either to the theory of storage where the risk premium depends on inventory levels, and thus on the slope of the forward curve, or to the hedging pressure where the risk premium is a function of hedgers’ and speculators’ net positions. These theories have been empirically validated in numerous empirical studies, all of which highlight that futures returns depend on the fundamentals of backwardation and contango.

In all of the empirical studies, the scheme employed to weight commodities within a portfolio is equal-weighting. The rationale for this choice comes from the fact that unlike in equity markets, there is no natural value-weighting that can easily be applied; its equivalent (production and consumption weighting) is hard to implement given how difficult it is to collect reliable inventory data. This paper proposes to relax the assumption of equal weights and to test whether weighting schemes emanating from the equity literature could be more profitable. These weighting schemes pertain to mean-variance optimization and volatility timing.

Our first contribution is to apply the weighting schemes of the equity literature to the commodity markets. In total, eight weighting schemes are considered; these can be split into an equal-weighting scheme, two optimization schemes, and five timing schemes. The equally weighted scheme is standard in the commodity pricing literature. The optimization strategies follow Markowitz (1952) and are based on mean-variance (MV) and minimum variance (MIN). When it comes to the timing strategies, three approaches follow Kirby and Ostdiek (2012) and define portfolio weights based on volatility timing (VT), beta timing (BT), and reward-to-risk timing (RRT); the other two are novel and based on tail risk as modeled via Value-at-Risk (VaR) and conditional Value-at-Risk (CVaR).

Our second contribution is to amend the aforementioned weighting algorithms so as to consider the specificities of backwardation and contango that prevail in commodity futures markets. We do that by adding a novel step to the weighting procedure that considers either past performance or the slope of the term structure of commodity futures prices as a buy or sell signal for each of the commodities present in the cross section at the time of portfolio formation. This method allows for the possibility of being either long or short while mitigating common issues such as extreme weights or artificial inflation of returns due to the self-financing nature of long-short positions.

Our findings indicate that the optimization-based and timing strategies perform well, as measured by a range of risk-adjusted return metrics (Sharpe and Sortino ratios, Maximum Drawdown, and VaR and CVaR). We find no evidence that returns from the investment strategies are compensation for liquidity risk. Our investment strategies are robust to transaction costs and the the choice of model parameters. We show two alternative methods of classifications, that nominate commodity futures for long or short positions based on momentum and term structure, exhibit similarly strong performance. We also show that our strategies are stable in various market conditions, from crisis to high growth periods. we document that these strategies are amongst the most profitable in the literature and show that common risk factors in the commodity futures market are unable to account for their positive and significant excess returns."


Are you looking for more strategies to read about? Check http://quantpedia.com/Screener

Do you want to see performance of trading systems we described? Check http://quantpedia.com/Chart/Performance

Do you want to know more about us? Check http://quantpedia.com/Home/About

Factors vs. Sectors in Asset Allocation Wednesday, 7 June, 2017

What is better - factor of sector investing? A recent paper takes a look on this question a offers an advice. Analysis is related to multiple smart beta strategies:

Authors: Briere, Szafarz

Title: Factors vs. Sectors in Asset Allocation: Stronger Together?

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2965346

Abstract:

This paper compares and contrasts factor investing and sector investing, and then seeks a compromise by optimally exploiting the advantages of both styles. Our results show that sector investing is effective for reducing risk through diversification while factor investing is better for capturing risk premia and so pushing up returns. This suggests that there is room for potentially fruitful combinations of the two styles. Presumably, by combining factors and sectors, investors would benefit both from the diversification potential of the former and the risk premia of the latter. The tests reveal that composite strategies are particularly attractive; they confirm that sector investing helps reduce risks during crisis periods, while factor investing can boost returns during quiet times.

Notable quotations from the academic research paper:

"Factor investing has recently become a huge success in asset allocation. But its supposed superiority over other portfolio management techniques has yet to be proven. To fill that gap, we lay down a challenge to factor investing by organizing a contest pitting it against a well established competitor, the classical industry-based approach to asset allocation. We compare the financial performances of factor-based and industry-based asset allocations in the investment universe composed of U.S. equities. We contrast the mean-variance performance of diversified portfolios made up of sectors with diversified portfolios composed of the five factors developed by Fama and French. We
duplicate all the trials for long-only portfolios (no short sales) and long-short ones (unlimited short sales accepted).

Our contest reveals no overall winner. In fact, we find circumstantial evidence of superiority for each style. The alphas of factors with respect to the market inflate expected returns, while sectors reduce risks through high diversification potential. Factor investing tends to dominate when short sales are permitted. By contrast, when short-selling is excluded, industry based allocation is preferable, especially for highly risk-averse investors. These balanced results lead us to conjecture that factors and sectors could be complementary investing styles, and that combining them should help enhance financial performance, at least under some configurations regarding short-selling and/or risk aversion.

Our empirical investigation suggests that composite portfolios made up of sectors and factors are particularly attractive under two types of circumstances:

First, during crisis periods, a mixture of sectors and factors largely dominates whichever style is the best standalone performer.

Second, moderately risk-averse investors will find it best to combine sector and factor investments."


Are you looking for more strategies to read about? Check http://quantpedia.com/Screener

Do you want to see performance of trading systems we described? Check http://quantpedia.com/Chart/Performance

Do you want to know more about us? Check http://quantpedia.com/Home/About

An Analysis of Momentum Behaviour in a Long-Term Friday, 26 May, 2017

A recent paper takes a look on a long-term behaviour of momentum portfolios. Related to all equity momentum strategies, mainly to:

#14 - Momentum Effect in Stocks

Authors: Ali, Daniel, Hirshleifer

Title: One Brief Shining Moment(um): Past Momentum Performance and Momentum Reversals

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2956493

Abstract:

Motivated by behavioral theories, we test whether recent past performance of the momentum strategy (Past Momentum Performance--PMP) negatively predicts the performance of stale momentum portfolios. Following periods of top-quintile PMP, momentum portfolios exhibit strong reversals 2-5 years after formation, whereas, following periods of bottom-quintile PMP, stale momentum portfolios earn positive returns. The difference in cumulative five-year Fama-French alphas for momentum portfolios formed in high- and low-PMP months is 40%. A value-weighted trading strategy based on this effect generates an alpha of 0.40% per month (t = 3.74). These patterns are confirmed in international data. These findings present a puzzle for existing theories of momentum.

Notable quotations from the academic research paper:

"A set of studies propose behavioral hypotheses to explain the momentum anomaly. An implication of some of these models is that the momentum phenomenon is a result of delayed overreaction to certain information shocks. This implies that a sufficiently `stale' momentum portfolio, where `stale' refers to a momentum portfolio formed at a lag of twelve months or more, will on average earn negative abnormal returns.

However, to our knowledge, no study has yet examined the conditional variation in the performance of stale momentum strategies, i.e., the performance of momentum portfolios in years 2-5 post-formation. One interesting possibility, motivated by the idea that investors chase past style performance, is that strong recent past performance of the momentum style will cause investors to overvalue new momentum portfolios, resulting in poor subsequent long-run performance of these portfolios. In this paper, we explore this issue by testing whether long horizon performance of momentum portfolios is negatively related to the performance of the momentum strategy in the recent past.

In particular, we examine the relation of stale momentum returns to a measure of the recent performance of the momentum strategy, which we call Past Momentum Performance or PMP. PMP is simply the return of a standard (12,2) momentum strategy over the preceding 2 years (24 months). Our basic fi nding is that momentum portfolios formed in high PMP months (months when PMP is in the top 20% of all months in our sample) generate strongly negative returns and alphas 2-5 years after formation. Strikingly, momentum portfolios formed in low PMP months continue to (weakly) outperform in post-formation years 2-5. Thus, the momentum reversal documented by Jegadeesh and Titman (2001) is strongly state dependent.

We explore a set of behavioral hypotheses for the strong dependence of stale momentum performance on PMP. One of our hypotheses is based upon style chasing.

A basic hypothesis is that the performance of the momentum style will tend to continue in the short run, so that after the momentum strategy has done well, it tends to do well again. The style chasing approach suggests that following high returns on the momentum style, owing to return extrapolation, naive investors switch into this style, meaning that they buy winners and sell losers heavily. This trading pressure reinforces the strong performance of the momentum strategy, and will temporarily cause better-than-usual momentum performance after the conditioning date if such return chasers arrive gradually.

This e ffect is driven by overreaction in the components of the momentum portfolio. In consequence, the returns on the momentum portfolio will eventually reverse. So after high PMP, there are on average negative returns to a stale momentum strategy of buying firms that were winners at least a year ago and selling firms that were losers at least a year ago.

In contrast, after low PMP, investors switch out of the momentum style. Heavy selling of winners and buying of losers induces underreaction in winner and loser returns. So after low PMP, this hypothesis implies eventual positive returns to a stale momentum strategy. Putting these two cases together, we expect reversal of momentum to be stronger as PMP increases.

Motivated by these ideas, we examine the relationship between PMP and the performance of stale momentum portfolios and fi nd a number of novel e ffects. We fi rst show that over the full CRSP sample, there is on average very little tendency of momentum to reverse after controlling for the value eff ect. This fi nding is in contrast to that of Jegadeesh and Titman (2001) who find, in a shorter sample, that equal-weighted momentum portfolios exhibit strong reversals even after controlling for the value e ffect.

Then, turning to our main result, we fi nd a strong relationship between PMP and long-run reversal of momentum - reversal is greater after high PMP. Speci fically, we rank the months in our sample into quintiles based on PMP and examine the performance of momentum portfolios formed in each category of month (i.e., for months in each PMP quintile) during the five years after formation. Stale momentum performance declines strongly and monotonically with PMP. In Quintile 1, instead of reversal, momentum portfolios exhibit weak continuation in post-formation years 2-5. In sharp contrast, momentum portfolios formed in Quintile 5 months lose 42% of their value over the next fi ve years. We call this strong reversal of momentum after high PMP the PMP eff ect.

table 1"


Are you looking for more strategies to read about? Check http://quantpedia.com/Screener

Do you want to see performance of trading systems we described? Check http://quantpedia.com/Chart/Performance

Do you want to know more about us? Check http://quantpedia.com/Home/About

An Example of Trading Model Design by Richard Olsen (Founder of OANDA) Friday, 19 May, 2017

A very interesting example of FX trading strategy created by Richard Olsen (Founder of OANDA):

Authors: Golub, Glattfelder, Olsen

Title: The Alpha Engine: Designing an Automated Trading Algorithm

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2951348

Abstract:

We introduce a new approach to algorithmic investment management that yields profitable automated trading strategies. This trading model design is the result of a path of investigation that was chosen nearly three decades ago. Back then, a paradigm change was proposed for the way time is defined in financial markets, based on intrinsic events. This definition lead to the uncovering of a large set of scaling laws. An additional guiding principle was found by embedding the trading model construction in an agent-base framework, inspired by the study of complex systems. This new approach to designing automated trading algorithms is a parsimonious method for building a new type of investment strategy that not only generates profits, but also provides liquidity to financial markets and does not have a priori restrictions on the amount of assets that are managed.

Notable quotations from the academic research paper:

"To summarize, our aim is to develop trading models based on parsimonious, self-similar, modular, and agent-based behavior, designed for multiple time horizons and not purely driven by trend following action. The intellectual framework unifying these angles of attack is outlined in Section 3 of source research paper. The result of this endeavor are interacting systems that are highly dynamic, robust, and adaptive. In other words, a type of trading model that mirrors the dynamic and complex nature of financial markets. The code can be download from GitHub [The Alpha Engine: Designing an Automated Trading Algorithm Code. https://github.com/AntonVonGolub/Code/blob/master/code.java. Accessed: 2017-01-04. 2017]

The Alpha Engine is a counter-trending trading model algorithm that provides liquidity by opening a position when markets overshoot, and manages positions by cascading and de-cascading during the evolution of the long coastline of prices, until it closes in a pro t. The building blocks of the trading model are:

- an endogenous time scale called intrinsic time that dissects the price curve into directional changes and overshoots;
- patterns, called scaling laws that hold over several orders of magnitude, providing an analytical relationship between price overshoots and directional change reversals;
- coastline trading agents operating at intrinsic events, defi ned by the event based language;
- a probability indicator that determines the sizing of positions, by identifying periods of market activity that deviate from normal behavior;
- skewing of cascading and de-cascading designed to mitigate the accumulation of large inventory sizes during trending markets;
- the splitting of directional change and, consequently, overshoot thresholds into upwards and downwards components, i.e., the introduction of asymmetric thresholds."


Are you looking for more strategies to read about? Check http://quantpedia.com/Screener

Do you want to see performance of trading systems we described? Check http://quantpedia.com/Chart/Performance

Do you want to know more about us? Check http://quantpedia.com/Home/About