The digital versus analog trading world - The switch away from digital

 

A key advancement in many quant models is the switch from digital to analog signals. What do we mean by the this? A digital signal is binary. It is either on or off. If the price is above a moving average, you go long. If the price is below the moving average, you go short. The decision is a switch. You can add more conditional switches but the basic model is still digital based on switching.

However, many models are switching to analog. There is no single on or off switch but a way of thinking that is more probabilistic. There is a greater probability from the model that it is time to move to long or short. Positions are scaled with likelihood. There is no one single answer but a view that it is more likely to switch. This is more of a Bayesian view for making decisions. An analog model will lead to a different form of trading because there is no single answer on what action to take.

Supervised ML approaches - Lots of choices

 From Angela Shi

It used to be that analysts would get a set of data, form hypotheses, and then use regression as the model workhorse. Those days are gone and that is a good thing. One of the major advancements from the ML revolution is a providing a broader set of tools to solve data problem. A quick look at the set of supervised learning tools shows the growing complexity of choices. These enhanced tools are especially important as data sets get larger and more complex.  

Are more tools always better? No. The new challenge is learning the unique features of these tools and determining when is the right time to use them. Now, quants must set up models, find data, pick the right tools, and then build the portfolio. The choice set is more complex and requires new skills. A manager's comparative advantage will be associated with making the right tool choice.

Who represents smart money? Understanding who drives return performance

 

There will be talk about "smart" money and "dumb" money. Investors will talk about crowded trades. There will often be stories about groups of traders driving markets. The flows will matter. The dynamics of the market have been given a lot of press but there has been limited thinking about who profits from all this trading and the behaviors of different investors and how they process information. 

A recent paper tries to solve this problem. See "Which Investors Drive Factor Returns?" by Morad Elsaify. In this work, the author focuses on the processing of information as imbedded in the risk factors. The paper looks at a large set of risk factors and then measures the behavior of different trading groups around these factors. Different trading groups will show different portfolios selection around risk factors.

The processing of market information can come in two forms or choices: the persistent of fundamental risk factors or choice of risk factors, and the timing of idiosyncratic risk around a risk factor or factor timing. 

The fundamental trading is associated with the selection of risk factors while changes in risk factors is associated with idiosyncratic risk or factor timing. Smarter money will be able to factor time and focus on idiosyncratic risk while less sophisticated traders at processing information will focus on factor selection. Hedge funds seems to take more factor timing risk while more passive investors will focus on factor selection. Hedge funds will make markets efficient and act like arbitrageurs for factor returns. They will buy cheap factors and sell rich factors which will be idiosyncratic differences away from long-term averages. While those with less skill will just focus on factor selection or the fundamentals and not the transitory shocks in risk factors. 

Traders need to use information about the risk factors to solve two types of uncertainty, the average pay-off of a risk factor and the transitory or idiosyncratic portion of a risk factor. The attention or detail to processing information about any risk factor will determine whether a trader will be a timer or factor selector.

This not an easy paper to read, but it is highly suggestive of how information is processed by different trader types and provides a good explanation for what hedge funds do and how they behave.

Mind the Momentum Gap - A Key Indicator

 

Momentum is one of the bets know and consistent risk premium in the finance factor zoo. The strategy us simple. Sort the returns of some stock universe and buy the top winners and sell the losers. It is not considered part of the five factor "holy grail", yet there may still exist some puzzling results. Foremost the times series of returns from this factor are highly variable. While consistent in the long-term, there are periods of poor performance.

An interesting piece of research has been published that tries and explain the time variation in momentum based on the dispersion or gap between the winners and losers. It has been called the momentum gap. See "The Momentum Gap and Return Predictability" by Simon Huang. 

The momentum gap negatively predicts momentum profits. This applies to both the US and international stock markets. If there is a one standard deviation increase in the gap, there will be predicted a 125 bps fall in monthly momentum returns even after controlling for other factor predictors. 

There are several conjectures for why this momentum gap effect may occur, but a simple explanation may be best. If there is a large gap, then we have a case where the winners have significantly outperformed the losers. We are at extremes. There is a wide difference in performance which suggest that the momentum or trends that are being identified have reach extremes. It is likely that the momentum effect will be reversed. This is consistent with the fact that the momentum effect exhausted over the longer-run. 

The factor price for risk changes through time and market extremes have a different risk profile. The extreme, or chance of reversal, view makes sense because this gap effect is strong for static portfolios but declines when there is dynamic monthly rebalancing. A new sorting will find a different set of long and shorts which may be break any buy or sell extremes.  A large gap suggests an overreaction in momentum which may lead to a reversal or convergence. A tight gap is an environment where new information could lead to more divergence. 

Mind the gap for momentum or trend portfolios.

"The inevitable may be certain, but it is not always punctual."

 

"The inevitable may be certain, but it is not always punctual." 

- Jim Grant.

I just heard this quote on a podcast with Jim Grant, one of the best financial commentators/journalists we have on markets. Old school can always be fresh. 

The biggest problem with forecasting is not getting the idea or concept right. It is getting the timing right. Whether it is inflation, a banking crisis, a recession or a market reversal, the signs are often early. The signs may be right in front of us, yet the time frame can be maddening slow. 

The variation on this theme is from Ruddy Dornbusch:

“In economics, things take longer to happen than you think they will, and then they happen faster than you thought they could.” 

Love and Hate with the Dollar - Just a Fad

There is political cost with being a reserve currency - Investors love to hate on you and look for a fall. When you are on top, everyone is looking for a fall because you cannot go any higher.  The dollar has gone through some rough patches but still has shown the ability to bounce back from negative stories.

An interesting story that has been given a lot of attention is dollar hatred expressed in Money: Inside and Out by Jens Nordvig in "A Brief History of Dollar Hatred". Nordvig makes the point that we have gone through several dollar hatred periods since 2000. During these periods of dollar hate, there is a strong narrative for why the dollar should decline only to see the dollar bounce back. I agree with the fact that we have gone through periods of dollar negativity but the reality, the dollar has shown resilience and moved beyond the hatred associated with a given time or story. 

The dollar hatred periods can be broken into three themes:

• 2004-08 - the great current account scare - The period when economists focused on the large US current account deficits which could only be solved through a dollar decline. 
• 2009-13 - The QE infinity story - Excessive monetary policy will lead to a dollar fall and the currency is debased from too much money in the financial system. Other central banks increased their money so on a relative basis the dollar was not debased as expected.
• 2021-present - Dedollarization by major countries who do not want to be hurt by potential sanctions. The dollar hatred is associated with US sanction hegemony.

The dollar has risen and fallen with changes in the macro environment. That should not be surprising. However, disequilibrium in the international finance environment will be solved through several channels. The price of the dollar is just one of those channels. 

In the case of the current account deficit, the rebirth of oil industry changed the current account environment. The QE infinity story was offset by higher growth in the US and the impact of QE in other countries. The dedollarization story is still playing out, but the limited convertibility of currencies for countries who want out of the dollar system is a major sticking point.

There is dollar hatred and there are issues which will drive the dollar lower, but dollar hatred cannot be looked at in isolation. The dollar has strengthened at key times because it is a place of safety. The GFC saw dollar gains. The pandemic saw flight to the dollar. EU problems after the GFC again made the dollar attractive. 

A negative narrative is just a story until there is a major price correction that is not being offset by other policies or price action.

Capital Decimation Partners and Capital Multiplication Partners

 

What makes a good hedge fund? Can you replicate hedge fund returns? These questions are not easy to answer when you think about replication strategies. You can create through bundling ideas a good hedge fund with a high Sharpe ratio, but there are hidden risks that may not seem obvious on the surface. We can think about the classic examples provided by Andy Lo with his humorous firms, Capital Decimation Partners (CDP) and Capital Multiplication Partners (CMP) as a case study on the drivers of hedge fund returns and replication. 

The returns of his fictitious hedge fund CDP are a combination of holding the stock index with selling out of the money puts for protection. You do well because you pick up premium every month until the time there is a large market downturn and then the pain begins. The hedge fund manager hopes that the size and frequency of a market decline just does not occur. Hope is not a strategy. The high Sharpe ratio comes because the tail event is not accounted for in the return to risk ratio. Increase carry in exchange for tail risk and you improve the Sharpe but create a different risk profile.

So, let's look at (CMP). In this case, the returns are generated through a switching model between the SPX and one-month Treasury bills assuming that you have perfect foresight on what will do better.  This timing model is like buying the SPX and with a put option struck at the price of the index plus the one-month bill return. Unfortunately, you cannot get those returns given the cost of the option and the issue not having perfect timing. We can only receive something less than the perfect forecast.  Nevertheless, we can use the thinking behind forming a hedge fund in theory to develop replication strategies through linear approximation. These will not be perfect, but it can be an alternative to buying hedge funds. 

Replication of a hedge fund can be tried through using several factors to find a linear fit, regression, between the hedge funds returns and a model. The alpha from the manager is the constant and residual or returns not associated with the linear regression. The results suggest that it is possible to clone the average return for hedge funds within a style category. Unfortunately, getting average returns is not what many investors want. Additionally, there is an error term with replication so you will not get something that will be close in the short run.

Looking at hedge funds as either an option program or a linear combination of factors is a good start to describe the risks when investing with these alternatives.

Precision in language across cultures - A problem

 

We are strong believers in language precision and have often talked about Sherman Kent and WEP, words of estimative probability. When using words to describe likelihoods and probabilities, there is not always agreement on meaning. This is what makes discretionary decision-making so difficult when placed in a committee structure. Someone can say an event is "likely", but there may not be agreement on what is the translated probability of the word. 

Recently, a friend sent the above chart on what words mean in other cultures for a given point on the normal distribution. It is humorous, but like most humor there is a grain of truth in the joke. Ask a person in a different culture and you will get some very imprecise meanings or translations. In Britain or Australia, one word covers everything. Think of this when you ask someone about recent returns or their monthly performance. "Not bad" or "It's fine" could mean anything. 

Frequentist versus Bayesian view of the world: A simple explanation

The frequentist approach forms an expectation from a sample of data. The Bayesian approach uses some knowledge to form a prior.  In the case of the frequentist, a sample of data is taken or observed and from that sample conclusions are drawn. In the Bayesian approach, there is a start with some knowledge, and data are used to update the prior knowledge. Conflicting evidence will lead to an update of priors. 

The frequentist assumes events are based on frequencies, the count, while Bayesian inference will draw on prior knowledge. The frequentist does not calculate the probability of a hypothesis. He accepts or rejects and is an absolutist. The Bayesian always thinks in terms of probabilities and reasons in relative differences.

The frequentist believes a parameter is not a random variable. The Bayesian says that a parameter is a random variable and measure likelihoods.

The frequentist will talk about a confidence interval, p-value, power, and significance. The Bayesian will use the term creditable interval, prior, and posterior. The frequentist will think about action to take, accept or reject hypotheses, and getting a right answer. The Bayesian will discuss opinions or prior beliefs and how they may be updated. There are no right answers only more or less likely answers.

If you are a trader, you are more likely to be a Bayesian and think about probabilities and priors.

from KDnuggests 

"I'm a pessimist, but that is no reason to be gloomy!"

 "I'm a pessimist, but that is no reason to be gloomy!" 

- the late Cormac McCarthy 

Cormac could have been a fixed income guy, but he was a physicist by training.  This phrase should be placed on the wall of every risk manager's office. Yes, you should look for risks, but that does not mean that the world is always going to come crashing down.

The replication problem in finance - You don't get what you thought

 

There is a problem with replication of finance studies on risk premia and trading strategies. Additionally, there is a problem of out-of-sample results not matching in-sample returns. You don't get what you think with most financial studies. There are a couple of reasons for this disconnect. One, the construction of the test was poor. Two, the articles published are only extremes from data mining. Three, risk premia, once observed are arbitraged away. Four, market behavior and structures change. Overall, the buyer needs to beware.

Investors are aware of research and exploit it so that future returns are lower. There is a life expectancy for strategies and once the cat is out of the bag and the general trading public knows the strategy, excess returns are quickly gone. The out-of-sample results also show that there are strong performance declines.

So, what is the slippage that you should expect from models? Studies have found that portfolio returns for a strategy may fall about 25% for out-of-sample work. The drop can be over 50% during the five years after publication. 

Quantpedia did a study of out-of-sample results and found that the return decline is universal across strategies. Testing a large set of strategies, they found an average drop of 33% out-of-sample and with the median drop over 40%. 

Nevertheless, there is a significant performance difference between in and out-of-sample results. There is a lot of variation. However, a careful analysis suggests that since there is factor momentum, there can be ways to reduce the out-of-sample problem. Hold the factor premium or strategy that is trending higher.

The noise bottleneck is a problem with decision making

The noise bottleneck is really a paradox. We think the more information we consume, the more signal we’ll consume. Only the mind doesn’t work like that. When the volume of information increases, our ability to comprehend the relevant from the irrelevant becomes compromised. We place too much emphasis on irrelevant data and lose sight of what’s really important. 

-Farnam Street 

Give me more information! The more data - the better the decision. No, this may not be the case. Cutting out the noise and focuses on less may be a better way of thinking about problem solving. Nassim Taleb in his book Antifragile, discusses the problem as the noise bottleneck. What is true in the absolute concerning data may not apply to the relative issue. As you consume more data, you may see the noise to signal go higher. More data creates a false confidence as well as stress for investors. For models, more data also create noise. Features will increase, but these features are often dynamic and move in and out of significance.

How do you solve this problem? First, realize that there is a signal to noise problem. More data is not always better. Second, preparing signals to streamline data use is appropriate.  Just because data is available does not mean it should be used. Three, accept that if signals cannot be counted and tested as repeatable events, it should not be used. Reading one-off stories as signals is not appropriate. 

See our past posting on signal and noise:

Kahneman on silence - Turn down the noise through turning up the algorithms

Decision noise reduction - This is the one thing investment managers should get right

The problem of distraction and overload - Filter the noise and stay focused

Noise versus bias - We focus on biases but it is the noise that hurts us

Pump down the noise - Decision silence

Greedflation ... Is there such a thing?

We have cost-push, demand-pull, headline, and core inflation to name just a few descriptors. We have also been exposed to shrinkflation, products at the same price but have less. Now we are hearing about greedflation from the chief economist of the OECD.  

We have heard this term before, but it now shameful to try and keep your margins the same during an inflationary period. If your costs are going up, shouldn't firms try and pass those costs onto customers? Customer demand may fall and the pass-through may not be possible, but it is in the interests of shareholders for managers to make the attempt to pass on costs. The customers will decide whether the margins can be maintained. Is this process greedy or just the normal behavior of businesses doing their job?

Machine learning as color wheel

 

Hat tip to Saul Dobilas for the great picture 

How do you choose the right technique for the right problem? This is a growing issue with machine learning because there are so many approaches to problem-solving. The first thing to do is classify techniques and the color wheel does a good job of providing a first pass. It is a good way to start to solve the technique choice problem. 

Tax receipts going south - what does it mean?

 

While the focus has been on the debt ceiling, you really cannot separate this from tax receipt issue. if the government spends today, it must come from either current tax revenue or future tax revenue which is delayed through financing. Borrow today but it will have to be paid in the future or it requires continual rolling the debt.

Tax receipts decline when you go into a recession. Less economic activity and there will be less income taxes, less corporate taxes, and less capital gains. The government must borrow to stay even with expenditures. The yearly change in tax receipts shows the stark effect on government financing. Obviously, any cut in taxes as growth slows only makes the picture worse. 

The current numbers are clear. Tax receipt growth has turned negative and that is not a good signal. There are periods when the change has moved negative but we have not had a slowdown but that was also a period that required the Fed to reverse course from starting raise rates and employ QT. While the recession risk forecasts have fallen, there are still signs of a slowdown.

Inflation and equity markets - Higher inflation is not better

 

Equity markets like low inflation regardless if inflation is rising or falling. On the other hand, equity markets do not like higher inflation. especially if it is rising. The relationships are not precise and there will be significant differences based on sector, but we do know that high inflation will hurt multiples with earnings compression, higher financing costs, lower expectations for valuation, and more uncertainty.  See "Which equity sectors can combat higher inflation?" from Hartford Funds.

Financial will be affected by the real rate and what the Fed will do. Energy and materials will do better if prices increase can be passed-through to consumers and businesses. Real estate should do better in high inflation, but again, there is an issue of financing. In general, the impact of inflation on equity returns can be very complex. While past inflation environments have been counted, we cannot make clear judgments concerning the current environment where inflation is falling but growth may also be declining. 

Break inflation into cyclical and acyclical components - A better story

 

Inflation can be broken into a cyclical and an acyclical component. Cyclical inflation components are those that are sensitive to overall economic conditions while acyclical components are those that are associated with industry specific factors. The cyclical components are sensitive to the unemployment gap based on the Mahedy-Shapiro method. See the San Francisco Fed piece "Cyclical and Acyclical Core PCE Inflation". 

We have seen a decline in the acyclical component since the Fed started to raise rates, but the cyclical components have been sticky and may only now be peaking. The Fed raising rates has not be successful at curtailing cyclical inflation given there has not been any impact on the labor markets. This evidence would suggest that more rate increases may still be necessary. Of course, the alternative story is that the Fed has done its job but there has been a delay in the economic response. Neither story is good for markets. In the first case, financial costs will have to rise further. In the second case, the Fed may have overreacted given a poor understanding of lag structures.

"The Future Is Real. The Past Is All Made Up."

 

"The Future Is Real. The Past Is All Made Up."

- Loggan Roy (from the series Succession)

We can manipulate the past through changing facts and interpretations. The past can be given many interpretations to fit what we want it to be. There is no one history, but the future is specific and cannot be changed or reinterpreted. It is a reality check.

We can make up investment narratives about the past but when those narratives are applied to the future, it will face the reality that our stories will either be right or wrong. We can build a model to describe the past, but the model will have to face the reality of the future. If it is wrong, there will be a price to pay.