Trading with signal and price impact uncertainty

 

When you want to make your model practical, you have to look at signal uncertainty and price impact. If you use any model, your one-step-ahead forecast may have some uncertainty because the average coefficient may not accurately represent the true coefficient sensitivity at a given time. Similarly, for any model there is going to be a prcie impact from trading. If the signla is weak then iapct uncertainy is not that important but as the siganl strengthens, the effect price imapct will become more important. See “Trading with uncertainty about signals and price impact.” These real-world effects are important if you want to trade a model.  

Expectation Bias and short-term Momentum

 

Machine learning can be used to predict analyst forecast errors. These forecast errors can predict cross-sectional returns and abnormal returns. There is an underreaction to fundamental news, which is amplified by overconfidence, sticky beliefs, and information uncertainty. This expectation bias can explain short-term price momentum in high volatility stocks. From expectation bias comes a rationale for trading patterns in price. See "Expectation Bias and Short-term Momentum". Past forecast errors have a critical impact relative to other features.

Attention versus earnings and momentum

 

One area of increasing research is the attention that is given to a specific market. Investors cannot follow everything, so there are different levels of attention. Given changing attention, there should be different levels of efficiency. What is found in the paper, "A Tale of Two Anomolies: The implications of investor attention for price and earnings momentum," is very interesting. Stocks that receive more attention exhibit greater price momentum and weaker earnings momentum. The authors make the following claim: investors pay less attention to earnings news, stock price underreacts, leading to stronger earnings momentum, but when attention is high, behavioral biases intensify, which fuels overreaction and price momentum. Depending on the type of attention, the effects will vary. 

Multi-agent LLM systems for profit

 

LLM can be used to develop autonomous trading systems. The key is to have agents mimic the workflow within a multi-agent trading framework that divides a trading system’s tasks. When you look at a deep decomposition of the tasks associated with a trading system, you will get better risk-adjusted returns. In the paper, “Toward Expert Investment Teams: A Multi-agent LLM System with Fine-Grained Tradign Tasks”, the researchers develop what they call a fine-grained system that looks at detailed tasks and instructions and compares it with a coarse-grained LLM system that does not provide detailed instructions to the agents. 

Across different stock portfolios, the fine-grained approach performs much better. It is notable that we can examine the effects of dropping specific agent tasks and assess their impact on the Sharpe ratio.

The results are clear - be detailed with the LLM tasks given to agents. The more specific you get, the better the results.

Mimicking managers for profit? Not so fast

 

Portfolio managers can be viewed as economic agents that have regularized behavior. Hence, if we can formulate that agent’s behavior. That is, we can mimic their behavior. If that is the case, we should then be able to generate a portfolio with similar returns. This is what is done in the paper, “Mimicking Finance”. The authors in this paper use AI and ML to extract and classify the behavior of economic agents that is predictable from past behavior. The agents may view that their behavior is novel. It is not. A large percentage can be mimicked, with approximately 70% of mutual fund managers’ trade directions predicted. The more history we have about a manager, the more likely we are to predict their behavior. Managers who oversee many funds exhibit greater predictability. But, it has been found that higher ownership leads to less predictability. 

So why do we need portfolio managers? We can just develop agents with different characteristics to do the job. There may not be that much special with many managers. This really shows the power of AI within the asset management world. However, we may have to step back with some conclusions.

Interestingly, managers whose behavior is more predictable significantly underperform their peers, and those who are the least predictable do better. If you cannot be typecast with your behavior, or you are changing your behavior, you will do better. 

The morning volatility uncertainty effect

 

There has been an increase in research on what happens between the close and the open, and from the open to the close, in asset markets. In the equity market, significant differences have been found between overnight returns and daytime returns. 

In the paper, “Early Birds Get the Vol; Morning Volatility Uncertainty and Variance Risk Premium”, the authors find that the volatility of the VIX or VVIX at 10 AM EST can strongly predict the next day’s asset return's variance. The authors mine the day to show that there is a significant risk premium that can be exploited. I have reservations about the results, but I’m interested in the finding that volatility shocks have a spillover effect on volatility the next day. 

Volatility shocks will have spillover effects that can be exploited. 

Can market forecasts front run information? The answer is yes

 

The foundation of market rational expectations is that a scheduled report cannot affect the firm’s stock price until it is reported. You can’t act on information you do not have. Yet, the paper “How markets forecast and ‘front run’ Information: Bayesian Market Efficiency” tells a plausible story that can explain some key market behavior.

The argument for this front-running is based on Bayesian statistics. Assume there is a Bayesian investor who, rather than waiting, infers what will be in a report before the release. Given this inference, the Bayesian investors will reprice the stock before the announcement. Yet there remains uncertainty about the content of any news that could lead to a rise in premiums, particularly regarding the precision and quality of the firm’s financial reporting. The premium can be either positive or negative. Hence, the stock price and the cost of capital will be impacted by how well investors can form expectations about a given financial report. 

It seems this already happens in markets. Investors form expectations about what a financial report will say, and those views are incorporated into the price. The Bayesian approach provides more structure to how expectations are formed. We are all Bayesians, so saying that we form rational expectations does not really tell us much about how expectations are formed in real life. Using Bayesian updating provides structure around our behavior.

Decoupling dollar and Treasury privilege

 

When we think about the exorbitant privilege of the US markets, the question should be divided into two parts. The Treasury privilege is associated with financing at lower rates than in other countries and with the dollar’s use across the information finance landscape. They are not exactly the same thing. 

The paper "Decoupling Dollar and Treasury Privilege" shows a difference in the convenience yields between Treasuries and the dollar. The convenience of the dollar is measured by the covered interest rate parity (CIP) between risk-free bank rates, SOFR, and the convenience yield of treasuries, which is captured by CIP deviations in government bond yields. There has been a divergence between these two convenience yield measures. 

The dollar convenience yield exists in the post-GFC period, while the convenience yield for Treasuries has turned negative, especially for longer maturities. The authors argue that this change in Treasury yields is related to the excess supply of Treasuries. This sends a clear warning that the US Treasury cannot finance debt at the current rate. 

Do LLMs make market more efficient? Yes

 

We have seen an explosion of new papers on how to use LLM in finance. Clearly, computing power can enable analysts to use vast amounts of data more effectively across companies. Many papers have shown that excess returns can be generated by using LLMs, but a more fundamental question is whether markets have become more efficient since the introduction of LLMs. This is the focus of the new paper “Do LLMs make markets more efficient?". 

The answer to the question of efficiency is yes. LLM availability eliminates 40 - 60% of post-new drift into the next trading day. LLM helps with short-horizon price discovery. The researcher used a novel way of determining this value-added by examining the effect of plausible exogenous outages of major LLM providers. When LLM providers are unavailable, there is more drift. 

Get on the LLM bandwagon, even though markets have already become more efficient with the use of this tool.  If you are not usng this tool, you are falling behind competitors.

False discovery rate in finance - Thinking out of the box

 

Aha! I found another risk premium. This has been the mantra of finance for well over the last decade, yet perhaps we should comment, "not so fast." While on the one hand, there has been an explosion of research finding new factors, what has been called the factor zoo, there has also been pushback by other researchers who have pointed out the issue of data mining. The argument of data mining is that conventional inferential frameworks are inappropriate when there are endless tests of different model specifications before the right one is reported. We will mine the data until we find the right result. Other researchers have pushed back against the data miner, with analyses showing that the posterior-expected alphas still exist even after the initial results were reported.

A new paper. "The False Discovery Rate in Finance: Identification failure and search-adjusted estimation" outlines how this problem can be solved. There should be a search-and-selection process for identifying factors. Knowing that process will help detect the false discovery rate, but if this search mechanism is unknown, there will be a greater likelihood of underestimating the true false discovery rate. 

Using some lower bounds calculated from their work, the authors suggest that most reported discoveries in finance are likely false. Ouch, that is a strong indictment and has strong implications for academic research and the work of quants who use that research to develop profitable trading strategies.  

Narrative and macro investing

 

An exciting area of macro research is the use of narrative to help explain the weekly movements in equity markets. This work is still in its infancy and seems to be taking several different directions. This work on narratives started with Robert Shiller and his research on narrative memes that may create bubbles. Another analytic approach has been the development of indices that attempt to measure risk by counting mentions of news events. It has expanded with the development of NLP and LLM models.

An interesting application has been developed using the GDELT database to create different narratives. These narratives are then used as input to a macro model that uses key economic data from FRED. See the paper, “Monitoring Narratives: An Applicaiton to the Equity Market" Using a set of key narratives developed around news themes, the authors find that added narrative information will increase R-squared and reduce error for a model trying to explain equity returns. 

This paper scratches the surface, but it does provide an interesting link between narratives and fundamental data to help explain equity returns. For all the focus on macro quant data, story-telling is still an important driver of markets.

Rationale for trend-following updated

 

The case for trend-following is well established, yet, given the fluctuations in returns during periods without a crisis, it is worth revisiting fundamentals and discussing the rationale for this hedge fund strategy. Meketa, the pension consulting firm, produced a white paper on the topic at the end of the year that provides some new insights. 

One, the dispersion between trend-following fund returns is significant. The difference between the best and worst can be over 25%, and the difference between the 25th and 75th percentile averages around 10%.

 

Two, the difference in dispersion will increase with the extremes in equity returns. When market dislocations are greater, there will be greater dispersion in returns. 

Three,  the Sharpe ratio can be smoothed and increased if an investor chooses a portfolio of managers. It may be hard to say what the right number is, but 4-6 seems to provide the benefits of smoother Sharpe and less dispersion in the return-to-risk trade-off 

Tax loss alpha is getting big

 

There has been an increase in stories about tax alpha and how this has become a big thing in the hedge fund industry. Hedge funds are not tax effciency. The active trading in many funds generates positive returns, but capital gains may be limited, so returns are generally treated as ordinary income. Managed futures will have some tax advanatges, but the general case is that invetsors should compare after-tax returns across strategies. 

The question is who should be generating the tax alpha - the manager or the investor. The answer is to look at some combination of both, There is the old adage by Buffet about the two rules of asset management: Rule 1 protect principal, and rule 2, follow rule 1. 

Of course, the top priority is for any hedge fund is generate return, yet, tax efficicny should be a goal that can provide improved returns without significnat risk. For those who have SMAs, the tax efficiency can be achieved by the investor and viewed more holistically. Wash sales, tax loss harvesting, and forms of tax defferral can all help reduce tax drag. As more "retail" investors get involved in hedge funds, the issue of tax efficiency will come to the forefront. 

Friction needs to be managed

Friction is like transaction costs. For business economists, this is often overlooked or treated as an afterthought, yet solving the transaction-cost problem is the critical driver of most business structures. The Friction Project, by two Stanford business professors, makes an interesting observation on the impact of friction on business success. What is interesting is that sometimes it is important to reduce friction to make decisions easier. Still, there are also times when friction, slowing things down, and making decisions harder can be beneficial.

The job of a manager is to cut through friction, another way of making businesses more efficient. For example, make memos shorter. Make meetings more focused. Friction adds to the grind of impediments to increasing productivity. Look for frictions and then cut them. Friction can also be used as a tool to make it more difficult to switch or decide. Yet, frictions may mask bigger problems. 

Transaction costs often seem so abstract, so I like the descriptive word “friction” as a better way to identify impediments to getting things done.

Managing the hard things - there is help from Ben Horowitz

 

Just read The Hard Thing About Hard Things by Ben Horowitz, a management book about experiences as an entrepreneur at tech start-ups. I often pick up these books for a quick read, hoping there may be a nugget or two on how to improve as a manager. The first part was not impressive, but Horwitz then goes on to offer practical advice on numerous topics managers face. How do you hire? How do you fire? How do you promote? How do you motivate? 

These practical tips are very good and can be implemented by any manager. Are these tips easy to use? No, making hard decisions is not easy, but Ben provided a guide on how one person has addressed these issues. The usual business management book will talk about cases, but readers want specifics like how to deal with well-defined problems. This book delivers. Be direct, be truthful, and do not try to avoid the hard decisions. 

The problem of bimodality and deep mometum

 

Momentum is considered one of the financial factors that shows consistency throughout time. Other factors come and go, but not so with momentum. This does not mean that momentum will work at all times. Additionally, momentum is subject to significant risks. It depends on the regime, with momentum profits greater during an expansion than during a recession. There is also a greater likelihood of crash risk with momentum trades. This is the premium investors are paid to hold assets showing momentum. Finally, it is found that momentum exhibits a bimodal distribution of relative returns. Past winners may be likely to persist, but there is also a greater likelihood of becoming losers. This U-shaped distribution creates specific risks when holding momentum stocks.

The paper, "Bimodality Everywhere: International Evidence of Deep Momentum,” explores this specific momentum feature by using a deep momentum technique (RET) to mitigate or exploit it. Deep momentum is a two-step process: first, neural networks generate probabilities of future return declines; second, stock performance is based on returns or the Sharpe ratio to form long-short portfolios. It is shown that using specific machine learning techniques yields significant improvements over traditional methods for forming momentum portfolios. 

There is a lot going on with this paper, so the devil is in the details, but it shows that machine learning can be used to improve over a simple momentum strategy.

Diversification and redundancy with trend-following

 

A core tenet of trend-following is to diversify. Diversify markets and diversify the signals. The diversification of markets is rather simple. Add uncorrelated markets to reduce overall volatility. The trend signal is diversified by adding different lookback periods. While medium lookback horizons usually perform best, research has shown that adding short- and long-term signals diversifies the signal set and reduces volatility.

The paper, “Revisiting the strucuture of trend premia: When diversification hides redundancy,” looks at the trend horizon signal more closely and finds that you can add too many signals. Specifically, if you employ both short- and long-term signals, you will likely render the mid-horizon signals redundant. Do not weight signals equally; use some optimization to balance across horizons. When optimization is employed, the mid-range is not useful, and a barbell approach is better. This is not the conventional wisdom often used when diversifying time horizons. A short horizon helps mitigate drawdowns, and diversifying horizons smooths returns across different regimes. The key issue is determining how many horizons are necessary, given that trend horizons are correlated. However, simple optimization does not seem to add value. The data needs to be examined more closely and manipulated. 

Fed Up - a reread on the politics of the Fed

 

Is the Fed political? Yes, to the extent that it is driven by its own self-interest and preservation. It is political in that there is significant inbreeding of thinking and culture. This is especially true of research. All well-trained economists who often think alike and are focused on publishing research on their chosen topics, rather than on driving better decision-making by the Fed. It is not that they do not want good policy. It is the fact that they cannot think of how to do policy that veers from the status quo. One could say that QE and zero-interest rates did not establish the status quo, but they were a solution that fit the macroeconomic narrative at the time. 

I reread the Book Fed Up to hear one insider’s view of the Fed. This is not written by the Chairman or a leading economist but by a Fed bank advisor. It was a little long and could have been more directed in its arguments. Still, anyone who reads this will appreciate that having well over 1,000 economists and multiple regional district banks is crazy and does not serve the country’s good or its policies. 

I am not arguing for lower rates, but I will argue that the Fed could use a good shake-up. 

Signal filtering for mean reversion trading

 

If you aren't a trend follower within the quant price space universe, then you are a mean reverter. A paper discusses how to filter these reversion signals, "Advanced signal filtering for mean reversion trading." Mean reversion is based on the simple concept that an asste's price will converge to some fair value. This fair value could be as simple as a moving average. The spot price may fall above or below this fair value price. To solve this problem, the authors develop what they call the local average filtering objective (LAFO), a low-pass filter that operates across different frequencies. LAFO examines the average residuals over a moving window to capture moving-average characteristics. This information can be used to measure or describe mean reversion. LAFO is an extension of the mean squared error. Machine learning can help process data to identify the method of reversion to the mean and mispricing in a time series. 

Can there still be dislocations that will cause mean-reversion not to occur? Yes, but by examining different rolling windows of residuals, there is a good chance of finding revision opportunities.  

The Investor, The Chef, and The Recipe Book

 

I have once again discussed with friends what is proprietary regarding strategy details. Should a manager give up details on how they model and generate returns? I understand that in a competitive market, no one wants to give up their edge or reveal how they generate returns. However, investors still need to understand how returns are derived because backtests and past track records offer only limited insight into the return-generating process. There needs to be a framework for understanding how returns are generated. 

I often use the analogy of a cookbook. There are many great cookbooks available in a bookstore. Great chefs write cookbooks and share their “secrets” in their recipes, often with detailed instructions for preparing the meal. Measurements are given. Ingredients are detailed, and step-by-step instructions are often provided for preparing the meal at home. You may follow all the instructions, yet you will often not get anything close to what you experience in the restaurant. You are frustrated and feel the effort was not worth it. So, what do you do? You go to the fine restaurant and pay the price. You are happy and appreciate all of the work that the chef has done to prepare the meal. All are pleased. Information has been provided with high transparency, resulting in a willingness to pay the price for a great meal. There is transparency, which makes paying the meal’s price more worthwhile.

The core of this story is that there is knowledge from books and theory, and then there is practical knowledge from doing. When you invest with a special strategy, the manager is providing their specialized practical knowledge that cannot be easily learned. The devil is in the details, and that is often the secret sauce. I appreciate the practical wisdom on how to gather the right data, manipulate it, and use specialized techniques to extract signals. That is proprietary, but the overall concept should be discussed. Practical knowledge is contextual and situation-specific, and not easily learned just by being told how something is done.

Unified framework for anomalies - all in the past month of return behavior

 

The daily return information factor (DRIF) is a new concept that helps explain many of the anomalies we see in financial markets. Instead of imposing or seeking a new risk premium, the authors of this paper, “A unified framework for anomalies based on daily returns”, examine the overall mapping of returns over the last month to make predictions about next month’s returns. The authors examine both the time ordering and the magnitude of returns to develop a forecasting framework.

A chronological vector preserves time ordering and captures short-term reversal dynamics, while a ranked vector accounts for magnitude effects. The DRI variable will combine these two vectors so that next month's returns are based on the beta of the time and magnitude vectors. A chronology dimension captures price pressure and liquidity effects, while ranking reflects investors' focus on extreme outcomes. These effects remain after controlling for other risk factors. 

Manufacturing employment and trend

 The US has talked about a resurgence in manufacturing, but the numbers do not suggest it is happening. Where is the surge? Construction, perhaps a small uptick after being flat. Manufacturing no change in trend. The same with mining and logging. Transportation has fallen off a cliff. Utilities are also below trend. Employment is always fighting productivity. We may be doing more with less, but workers are not seeing the improvement.  

China - the electrostate

This chart from Ember has me thinking about how to classify China. I am not thinking about the politics or the finance, but China as an innovative state. The future of geopolitical power is centered not on the military but on the ability to bend others’ lives through technology. This does not dismiss the political, but we have seen politics devoid of innovation and technology, and its scope is limited to coercion. Technology drives hegemony.

 

Is AI a stochastic parrot?

There’s a well-known phrase that AI is not true intelligence, but just a “stochastic parrot”. But human beings do quite a bit of parroting in conversation, as it turns out. So maybe the comparison is a little more complicated.

Six Questions on the Value of Humanities Research

A Podcast Interview with Chris Yeomans, Justin S. Morrill Dean of Liberal Arts, Purdue University

I found this phrase thought-provoking. We all parrot what we hear from others. Perhaps most of our conversations are parroting what we may have learned from someone else. AI is just a sophisticated parrot because it can search through more work than we can. AI, like us, does this with noise. It is a stochastic process. 

So what makes us or AI more than a parrot? The difference is what we can call creativity. It is rephrasing or making extensions or narratives that are not just a repeat of what we have heard before. The reordering of fact. The prioritizing of data is a part of the creative process. This is done by AI, so it is more than just parroting with noise. However, we should be aware that AI is creating something new through this reordering. If we ask the questions differently or use a different algorithm, we will create something different.

Commodity shocks and financial markets

 

The first quarter is a commodity shock quarter. We had the gold and silver bubble and it bursting at the end of Jnauary. That seems like ancient history versus all the uncertainty from the war in Iran. The broad market is off more than 5%, yet the energy sector is up over 35%. More importantly, the price of oil is up over 50%, and gasoline futures are growing by 60%. There is a clear link between oil price shocks and financial markets. It is a one-two punch to equity and bond markets, and we have strong evidence of its effect across decades, dating back to the shocks of the early 1970s. Most of these large shocks are self-induced through violence. 

Beyond the magnitude of the shock, the key issue is the time required to return to normality. Short-term shocks can have a strong impact on short-term returns but then reverse quickly. The longer the shock lasts, the more likely it is to have a real effect on growth and inflation. The effect on growth is simple. An increase in energy prices is a tax on consumers and production. However, the US economy is less oil-price-sensitive than it was in the 1970’s, so it is hard to use this period as a control or base case. For inflation, the impact is closely tied to the actions of the central bank. An oil price shock is a relative price change, not an increase in the general price level; however, if the Fed lowers interest rates to support the economy, this price change can trigger an inflation surge.

All eyes are on whether this conflict will be prolonged, and with each day it continues, this short-term shock will be revised into a larger, economy-wide recession-inducing crisis. For many in the emerging markets, energy shortages are real and already disrupting growth.

Reducing anxiety choice overload

When there is high uncertainty, there is anxiety in making a choice. Who wants to mkae a bad choice? The issue is then trying to reduce the level of anxiety to allow for better decision-making. Some simple rules can help reduce anxiety. 

Minimize the options - If there are fewer choices, it is easier to make a decision. One way to reduce the options is to categorize the choices. Allow for one option within a grouping. For example, if you are worried about an energy shock, break down the choice between risk-on and risk-off, and within each category, allow for two options.  

Stick to what you know - If there is an option that is not well known to you, then drop it from the choice list. Active decision-making may not be the time to learn about new options. It can be an expensive education. 

Do not regret wrong decisions - The anxiety of choice is based on the regret from making the wrong decision, so another simple way to reduce anxiety is to think about minimizing regret. What is the downside of making a wrong choice? It is important to remember that there are acts of commission, the actions we take, and acts of omission, the actions we don’t take, and regret is greatest for commission. Take the time to ensure that your actions are well-grounded to reduce regret.