Trend-following and improving portfolio construction - Know your correlation matrix

 

Trend-following is more than just finding trends. Portfolio construction matters. Research has shown that positioning based on volatility will help with portfolio construction (see "Trend-following with and without volatility scaling - Two different worlds"); however, accounting for correlation will provide even more value especially if markets move to correlation extremes like what we have seen in the last decade. Accounting for correlation adapts allocations processes to incorporate the marginal contribution to risk and allows for full implementation of risk parity principles. 

Many investors discuss risk parity without understanding the dynamics or benefits when these techniques are used. I am not stating that all risk should be equalized nor am I saying that investors should not include their trend views. Using all of the information in the covariance matrix will allow investors to incorporate changes in volatility but also the changes in the relationship across markets.  In a good paper, "Trend-following. risk parity, and the influence of correlations", Nick Baltas  explains the nuances of risk parity when applied with trend-following futures markets and shows how correlation extremes harmed those who just focused solely on volatility positioning. 

The paper focuses on seemingly minor but critical change in portfolio structuring and position sizing. A normal approach to position sizing is to use volatility parity which positions solely on relative risk. Positions can be set to equal risk through using an inverse to market volatility; however, this improvement does not account for correlation between markets. Volatility parity is suboptimal especially as markets become more correlated. Including correlation centers the problem on the contribution to risk. This slicing of contributions to total risk is the essence of risk parity and will more efficiently allocate exposures through using covariance terms. 

We will dispense with the difficulties with measuring correlation which is a much more complex topic. Baltas looks at the impact of employing volatility parity, naive risk parity, versus full risk parity which is equal risk contribution. This research finds that the difference between volatility and risk parity seems small when correlations between markets are low and do not move to extremes as seen in a crisis or if there is a single market driver, but there is a world of difference in the post-GFC period when correlations have increased and moved to higher levels. 

In the post-GFC period there is a significant difference between the choice of position sizing. Notice the significant differences in Sharpe ratio between correlation regimes. The Sharpe ratio can increase by a factor of 3 when in an extreme correlation environment. Not accounting for correlation will place a drag on performance and change the Sharpe ratio even if the same signals are employed. By this time, most firms have tried to account for risk contributions; however, it is important for investors to appreciate the value of these portfolio construction improvements.

Working on better signal extraction is always critical but improvement in portfolio craftsmanship is almost as important. A good set of signals can be masked when there is poor portfolio construction. A good model can be turned into a better model through accounting for the correlation and cross-market linkages. 

Hat tip to Darren @reformedtTrader for noting this paper.

Factor returns and the economic cycle - Performance will flip on changes in macro expectations

Factor returns move with the economic cycle. These factor relationships with the economic cycle are well-documented, so if the economic cycle changes or expectations about economic growth change, there will be changes in the relative performance of factors. This can help explain some of the factor switching that was seen in the spring, summer, and fall of 2020. Great upheavals in the business cycle will not only lead to changes in the market risk premium but significant shocks to factors returns even though they are often expected to be less volatile that market returns. 

Recent research show shows these relationships are more complex yet potentially valuable for global macro timers. Factors are forward-looking, that is, a factor may anticipate changes in the economic cycle. They embed expectations about the cycle; however, these relationships are often unstable. (See "Do Factors Carry Information About the Economic Cycle" by Marlies van Boven at FTSE Russell)

When economic growth is divided into three regimes, there are strong differences in conditional factor returns. Clearly, market returns are significantly higher in high growth periods over low growth regimes. On the other hand, quality and momentum factors do very well during low growth periods.  

These investment factors also tell us something about future economic growth. Size returns have a positive prediction on growth while quality has the opposite sign. On the other hand, value and momentum have less predictive power concerning future GDP. Nonetheless, value has additional predictive information beyond market returns. These market signals have actually strengthened since the GFC as noted in the second conditional panel. 

This evidence can be further sliced with inflation to provide a four regime world of expansion, slowdown, contraction, and recovery. Investors want to avoid value and hold momentum factors during an expansion, and avoid size in a slowdown and contraction. In a recovery period, holding market risk and momentum will be rewarded. Still, these rules are not hard and fast especially when looking at the post GFC period. 

Markets, which include the performance of factors, are forward-looking and provide expectation signals. They can anticipate economic expansion and slowdown, so the global macro trader who wants to engage in factor rotation has a tough job. The macro manager has to form economic cycle expectations and determine whether those view have already been discounted in factor returns. The gains can be significant, but these relationships are volatile and require strong opinions no different than those needed to forecast the overall direction of the market. 

Commodity super-cycle fever is here but shorter-term considerations should drive investment decisions

There were many reasons for strong performance across commodity markets in 2020, albeit equities showed better returns versus commodity indices. There are also good reasons for continued commodity support in 2021, but super-cycle fever may be premature. The chart below is just one example of how some analysts have posed the super-cycle argument.

 

Recoveries in commodity prices offer investment opportunities but these upward price trend may not be the same as a super-cycle. Commodity super-cycles are actually longer than the normal business cycle and are caused by a combination of events:

• Weak investment in production especially for extraction commodities which cannot be easily reversed in the short-run;
• Excesses in demand caused by long-term secular trends in growth that usually involve demographic shift such as the ascent of China as an economic engine or the overall growth in emerging markets;
• Shocks that disrupt shorter-term global supply chains but also have impact across more than one growing season; 
• Loose credit conditions for an extended period as signified by a declining dollar and the desire by investors to hold real assets as inflation protection. 

Commodity markets, at this time, look more like the strong price reversal after the GFC. The commodity index reversal after the crisis never reached the pre-GFC highs. In the GFC case, commodity markets fell significantly in 2008 only to reverse with strong gains in 2009 on lower rates and a rebound in growth. However, the market then fell into the long down cycle that may have only shown a recent bottom. 

The four conditions for a super-cycle are pointed higher, but business cycle considerations are still paramount. Credit conditions are easy and inflation expectations are growing albeit not showing excesses relative to old targets. Investment in extraction firms has fallen but production constraints have not tightened. The China - EM story is still strong however growth will be lower although on a larger base. The demand wild card is continued unprecedented fiscal expansion in developed markets. Finally, supply shocks were present which makes for some demand/supply deficits for selected commodities   

A super-cycle is good opportunity for the long-only passive commodity investor with a long investment horizon but mixed for the trend-follower who will often make trading decisions based on a few weeks of data. The trend-follower can ride a super-cycle wave but more importantly should focus on the business cycle supply/demand shock opportunities.  

Dispersion in CTA returns and small managers again lead with strong performance in 2020

Averages don't tell a good or complete story for CTA's in 2020. The average return for a set of managers reporting to SocGen Prime Services in their Nelson Report show positive performance but not what would expected given the size of market dislocations. We looked at large managers above $400 million in AUM and who also reported December numbers. This includes the generally favorable trading for the end to the year. The average return was 2.69 percent, but there were some strong winners with returns above 10% for the year. Dispersion was with a range of over 30%. 

When you slice the entire dataset based on performance for all CTA and quantitative macro managers, the numbers become more interesting. The average is 4.27 percent return which is almost 60% higher than the selected large managers. If we look at the top decile of performance, the 2020 average return is 30.96% or over 10 times higher than the returns for the large managers. Only 8 of 23 (35%) managers in the top decile are above $100 million in AUM and only 4 of 23 (17.5%) are in our large group category with AUM above $400 million. This top decile has a standard deviation that is about 40% higher than the large group. There is more risk-taking, but returns are significantly higher.

The small managers proved to be nimble in 2020 and again suggests that investors should not be focused on just the largest managers for return generation. Nevertheless, the investor has to be careful about taking on the added business risk from a small firm which may not have the same set of controls and infrastructure versus larger managers. Of course, the real question is whether these high performing managers will show persistence or just be one-hit wonders. 

Resilience engineering - Concepts that can be applicable to asset management

 

Some important concepts in other fields of study often provide useful insights for asset management. For example, there is a clear trade-off between optimality and brittleness associated with engineering problems. The best optimized solutions may not accommodate large shocks, frequent change, and uncertainty. A model may break after the first time conditions change from the period used for testing and training. An over-optimized model may be fragile. For quants, this is often the problem of overfitting. A system that is overfit to the past will provide a great result on past data but will deviate from expected performance or break once there is new information or there is a change in market sentiment not seen in the past. 

Investors need models or processes that are resilient. Nassim Taleb would call this the property of being anti-fragile, but his idea or term has not gained acceptance as a structural requirement from the viewpoint of the model builder. 

Erik Hollnagel, a Scandinavian expert in the fields of resilience engineering, system safety, and cognitive systems engineering, developed the key features or cornerstone concepts behind resilience. Resilience should have relevance for all investment and hedge fund management. A resilient system or model should have the following four skills: 

1. the ability to react appropriately when new information is encountered;
2. the ability to effectively monitor the environment in order to identify the current regime and any shift to a new environment; 
3. the ability to anticipate change in regimes or in responses to market events;
4. the ability to learn from mistakes and learn from the environment faced as new information is acquired.

A resilient model or system for making decisions should have some common properties:

1. Learning to live with uncertainty - uncertainty is not feared and a point of failure but addressed in a consistent manner;
2. Maintaining internal diversity - there is not just one solution or approach but a combination of approaches that can be reweighed based on changes in the environment;
3. Combining different types of knowledge - single paths or threads to problem solving will create brittle systems so alternative data and points of view are incorporated for flexibility;
4. Create opportunities for self-organization - some form of information organization that weighs factors that impacts predictions.

Investors should ask two simple questions. Do your models show resilience skill? Do your models have resilient properties?

The principles of resilience can be incorporated with how  research and modeling is approached. New data can be tested. Models can be reviewed and trained against different environments. Models can be developed to identify regimes to support switching of information weights. Processes can be developed to prepare for failure as opposed to reacting to model failure after the fact. Perhaps holding to consistent model specifications is the right answer, but resilience engineering will always be testing this assumption in order to prepare for the worst.  

Fight model overconfidence by having a threshold for failure

 

"Overconfidence Could Be Investors’ Biggest Mistake" - Richard Thaler, Behavioral Economist, University of Chicago Professor Barron's December 2020

There can be overconfidence with models. Markets change and sticking with one model in the face of changing conditions can lead to underperformance. A model that has worked in the past may not always work in the future. No method for adaptation is a recipe for model overconfidence. There is no acceptance for failure.

All models will fail or have periods of poorer performance. There is a frequency of success and while it may be stable and high, it can still be subject to periods of underperformance. Hence, there needs to be consideration on when the researcher should throw-in the towel and make changes. The modeler and thus their model can be overconfident. To not consider failure and the need for change is no different than the discretionary trader who has subjective beliefs that do not change or match reality. 

The question is when model change should be considered. Certainly an ad hoc approach for model change based on a run of failures is the most frequent approach but it does not take advantage of advances in data science which focus on prediction and success based on a defined fitness function. Fitness can have a threshold which if breached represents failure. Action can be triggered by predetermined criteria. This can range from the simple, a maximum drawdown threshold to more complex criteria which are regime based. The important issue is having a plan for when you lose confidence in a model. 

The center of gravity for futures trading is moving to China

 

Metals market pricing has moved to where the metals are being used, China. Copper price discovery is centered in China given there is so much demand for the metal in one location. The flow of grain price discovery is moving to where there is the highest demand, China. There is even a new pig futures contract in China. There are active futures contracts in China with significant volume and open interest in base and industrial metals and agriculture markets. 

Price discovery for commodities will occur at the center of commercial mass for supply or demand. The Chicago futures exchange became successful in the 19th century given the grain production in the Midwest and the growing demand across the United States. The London futures markets in metals and tropicals became successful given the high local demand for these goods and the city being a trading and banking hub. The same can be said of metals and cotton trading in New York. Chicago, New York, and London futures were the nexus for banking and trading between producers and end users. A natural place for futures trading.

There is no guarantee of trading and exchange success associated with one location, but cash activity creates an environment for risk management products, hedging, and speculation. Location economic find resource allocation and synergies with trading and banking skill creates opportunities for network advantages. 

There is often a first mover advantage with futures trading and the banking regulatory environment has to be right for active trading, but as more commodities are actively bought and sold in China, the risk management apparatus will follow regardless of the advantage of existing exchanges. 

There is nothing inherently good or bad with this transition, but it will mean that trading rules, oversight, and behavior will change. These structural changes will create uncertainty as the process of creative destruction strikes futures trading, but it will also lead to trading opportunities as these markets become easier to trade for all.   

Rational Ignorance and Asset Management - Trend-following signals information attention

Trying to keep up with all of the information and news generated every day is a daunting task. There may not be enough hours in the day to even assess all the analysis that is generated let alone doing independent research. There are many questions that should be asked to determine how to allocate time reviewing information. 

How many analysts should be followed? How many strategists should be read? How much time should be spent reading news? What if important news is missed? What if what news highlighted or studied proves inconsequential? These relevant questions which impact our knowledge or ignorance will affect performance. Some information will be used and while other information is going to be discarded or ignored.

Some information or news is more important than others. If it is repetitive, it can be measured, tracked , weighed, and followed, but a lot of news is new and infrequent. Allocating time and effort should be spent on unique information which can often be more impactful because it is a surprise, yet the majority may just be noise. The cost of being informed or ignorant is important. 

For example, time spent on knowledge acquisition has been an ongoing area of voter study. The impact of a single vote may be small versus the effort to become informed. This cost is a reason for why voters show "rational ignorance". Voters may remain ignorant or uninformed because the cost of being informed may outweigh the impact of their vote. The cost of being wrong or right may be much greater for an investor, but the calculus of deciding whether to be better informed or stay ignorant is still an important decision. 

Classic economic orthodoxy states that an investor will conduct a cost/benefit analysis with respect to the acquisition of information, yet this does not answer the question of how rationally ignorant or informed an investor should be when faced with uncertain information and outcomes. There needs to be tools to fight the effects of ignorance and help focus attention. 

So how do investors fight ignorance or engage in successful rational ignorance?  We know that trying to learn everything will not work. The simplest solution is to invest passively and hold the market portfolio. This will allow an investor to receive the market risk premium without having to acquire information. A perhaps better solution is to not acquire information but measure the impact of new information in the market through trends. 

Trends serve as a signal of impact of information. Trends are the aggregate effect of the action of investors from their assessment of information. There is the assumption that others are information gathers, but the basic premise is simple. Use trends as a highlight for new information activity. If there is no new information, there should be limited trends. If there is a trend, it signals there is information activity that is worth our attention and focus. 

For some, the trend identification is enough information to act. There is no need to find the primal information. For others this serves as a catalyst for further inspection. By definition you will be late to market response to information, but it provides a low cost method for sorting through large sets of data. Trend-following can break the cost of information ignorance.

Trend-following with and without volatility scaling - Two different worlds

Trend-following seems generate positive returns across all market sectors and over long time periods. There may be stronger and weaker periods of performance, but the long-term historical record is trend favorable. However, there is some conflicting evidence with how successful trend-following is measured and structured. Not all trend-following is alike, or more importantly the historical success is partially an artifact with how the tests are conducted. 

Care has to be applied with determining how to best form a times series momentum or trend program. For example, the exhaustive study "Time Series Momentum" by Moskowitz, Ooi, and Pedersen found positive value from trend-following when structured with volatility scaling. A subsequent paper "Time Series Momentum and Volatility Scaling" by Kim, Tse, and Wald analyzed similar data and found that trend-following is no better than buy and hold when there is no volatility scaling. Their criteria for trend-following success is the measurement of alpha from a multi-factor enhanced Fama-French-Carhart model and a comparison with buy and hold alpha for a wide variety of markets versus a multi-factor model. This is not the same as saying that trend-following generates positive returns. 

How risk is managed matters, and the use of leverage is critical with futures trading. Not using the leverage in futures diminishes performance and alpha potential. The figures below show relative performance between scaled and unscaled alpha versus a buy and hold position. This research also finds that when buy and hold also scales volatility the gains from time series momentum is diminished, and the performance of cross-sectional strategies while positive will show differences in relative performance when there is volatility scaling. 

This has been an ongoing issue for discussion with trend-following managers. Do you volatility scale or not? Use risk parity or not? I see both sides of this argument and have been of the view that sizing based on volatility is useful but if overdone it can conflict with the goals desired by a trend model. 

Not surprising, the value-added for investors is in the details of how risk is managed, markets are bundled, and leverage effectively used. Investors should pay a premium for portfolio management expertise. The discovery of trends is critical but the true differentiator among trend-following firms is the management of risk. 

The ultimate goal of trend-following is to provide positive convexity versus a target benchmark. Convexity gains are focused during periods of market dislocation and not as obvious during long periods of return analysis attempting to measure alpha. Risk adjustments that diminish portfolio convexity harm this core goal, yet volatility scaling will improve overall portfolio characteristics in the long-run. 

Commodities will be subject to weather shocks (La Nina) - Opportunities for divergence

The Climate Prediction Center's Oceanic Nino Index shows cooler waters in the Pacific Ocean which is having an impact on global weather. The index is headed to decade lows which will translate into strong moisture and temperature differentials. There has already been an impact on some commodities, but this may spillover to the spring. 

Supply shocks will lead to stronger agricultural trends because supply cannot be easily be replenished. Agriculture supplies are inelastic in the short-run. We cannot predict the strength of any trend, but odds for a price divergence are better when potential shock conditions strengthen. The current forecast from Columbia University's International Research Institute for Climate and Society shows some potential weather extremes. 

That said, past La Nina events have not had an appreciable impact on, for example, Brazilian crops in the past, see cropprofit.com feature on ENSO cycles. This is the challenge for speculation, there is a risk set-up situation, yet the opportunity may not present itself until prices start to move. 

Always about the uncertainty - The good and bad in business survey data

Macro market valuation has to consider uncertainty with any longer-term assessment. The Atlanta Fed Survey of Business Uncertainty provides good expectational information that can help with this assessment. The survey asks business managers their 4-quarter ahead expectations for sales and employment growth as well as an assessment of business uncertainty as measured by the 4-quarter standard deviation for sales and employment. December number were recently published and contain both good and bad news. 

The good news is that the smoothed value for sales revenue and employment are still moving higher and the business uncertainty is falling for both series. This data suggest that businesses are looking through the current COVID case increases. The bad news is that sales revenue will still be below the average for the three years before the COVID shock. Employment growth looks to be at or above the prior three years. However, uncertainty, albeit below the high earlier in the pandemic, is still at high levels and suggests that businesses are still not sure of the environment a year from now. 

While monetary and fiscal policy are providing tailwinds for markets, the direction of the real economy is still far from certain. 

Financial asset overshooting and the disconnect with the real world

Are current financial markets irrational? Are equities and bonds in a financial bubble? These are some of the main questions to be discussed at the beginning of this year. There are many ways to address this question, but a continued theme is the disconnect between financial assets and the real world. We are not going to answer directly the initial questions; however, I will shed some light on the disconnect between the financial and real world from some key recent research on the topic. 

The model of a leading macro researcher Ricardo Caballero with the help of Alp Simsek provides some useful insights on this topic in their paper "Monetary Policy and Asset Price Overshooting: A Rationale for the Wall/Main Street Disconnect". They take the simple argument that the Fed wants to minimize the output gap that exists in the economy. This output gap will be affected by increases in wealth through financial prices, with a lag. If the Fed can push financial prices higher than normal, overshoot, then the gain in extra wealth will help reduce the output gap and provide a faster path to recovery. The words trickle down are not used, yet this is the direction of this work. Push up prices and the output gap can be closed. Push up prices early and the Fed can preempt output gaps. As long as the link between the financial and real world is lagged there will be a disconnect between the financial and real economy.  

Financial prices are something that the Fed can control immediately through their policy activities and raising these prices can support higher aggregate demand. The Fed is acting on multiple fronts but allowing for inflated asset prices is a clear policy choice priority. 

This conclusion should not be a surprise. It should be viewed as a tailwind for all asset allocation decisions. Because this works with a lag, Fed activity today can increase financial prices immediately under the hope that Main Street will be positively affected tomorrow. Higher asset prices with a weak economy can occur and be the result of policy action and not irrationality. The end result is a disconnect between Wall Street and Main Street. The trickle down affect has timing differences.     

The "risky shift" effect - A problem with group decisions and a reason for preferring models

 

Many studies of group behavior shows that groups prone to risk behavior will take more risk after discussion within the group and those that are prone to caution will get more cautious after discussion. The group will reinforce their behavior in either direction if individuals have similar views going into a discussion. The crowd moves to extremes. This is called the "risky shift" effect or phenomenon.

There is strong meme concerning the madness of crowds and crowd behavior. History is replete with examples, yet less has been written about the crowd behavior moving to more conservative positions. Additionally, the crowd stories focus on group frenzy, yet the move to extremes can occur even in a hushed board room or the conference room of asset management firms where a small group may be assembled. The impact of this effect has really not been studied much within asset management groups.

The reasons for this phenomenon are varied. Some suggest it is associated with diffuse responsibility, more confident participants persuade others, and greater social status of following the group. An individual's risk sensitivity shifts within a group. This is group think in a risk dimension. 

Forget the wisdom of crowds when everyone is thinking the same. Put the group in a room with no diversity and watch the feedback loop go to the extreme. The move to extremes is not just for the positive but also for the negative.

Quantitative models have an edge because they will not suffer from the risky shift effect. There will not be a move to risk extremes because the risk is programmed into position-taking. You will get exactly what you want, no more or no less.  

Small and large firms behave differently over the business cycle

The business cycle affects firms differently based on size. Large firms as measured by the top 1% in size are less sensitive to the business cycle than small firms. However, since these larger firms have high and rising concentration, the impact of small firms is less on the aggregate economy. Small firms may disappear and will not be missed by the economy as a whole as industries become more concentrated. Yet, the impact of small firm failure and growing concentration is not to be dismissed. The impact of economic shocks on firms of different sizes is an important area of research and understanding this affect will be important for those investing in small firms.

Unfortunately, the standard financial accelerator argument that the cyclicality of firms by size is based on financing constraints and frictions does not seem to be as clear-cut as measured by a recent study in the American Economic Review, (November 2020) "Small and Large Firms Over the Business Cycle". While financial constraints may have an impact on firms of different sizes, the relationship between size and financing may be less clear. This is important because current monetary policy is supposed to be geared to helping small firms, yet the premise that credit is the main problem is not substantiated in a close examination of the data. The authors are careful with their analysis. Financial constraints and credit channels may be important but the differential between small and large firm effects is more complex than described by financial accelerator models. 

The negative impact of a recession on sales based on size is significant, but the authors find that the difference in cyclicality of these firms is not based traditional proxies for financial strength. Policies that try and target financing for small business may not be as effective as thought from earlier research which focuses on the financial constraint channel. What seems to be important factors on the size effect are economies of scope and customer capital. Small firms, by being more focused, are more vulnerable to an economic downturn. More concentrated firms by region and product are more susceptible to economic shocks regardless of their financial situation.

2020 commodities - Almost all markets above average price for the year

The global recession has not stopped many commodities from having a good year. There is not a bubble in commodities like financial assets. Commodity returns show strength in demand and the impact of some localized supply shocks. All commodity prices are above their 2020 averages. The only exception is the natural gas market which is seeing the effects of milder weather along with lower overall energy demand. 

The only sector that has stayed somewhat rangebound after the March shock is energy. Precious metals have gained on overall market uncertainty and higher inflation expectations. Industrial metals have done well given the strong rebound in China. Agriculture markets both in grains and tropicals have performed well. Food demand has maintained and some weather shocks have provided added lift. 

Any strength in 2021 global growth will allow for continued increase demand in all sectors. Given current prices do not reflect speculative excess, commodity markets offer investors good diversification and upside returns.