HomeInformation EfficiencyTrend following: combining market and macro information

Trend following: combining market and macro information

-

Classic trend following is based on market prices or returns. Market trends are relatively cheap to produce, popular, and plausibly generate value in the presence of behavioral biases and rational herding. Macro trends track relevant states of the economy based on fundamental data. They are more expensive to produce from scratch and generate value due to rational information inattentiveness. While market trends are timelier, macro trends are more specific in information content. Due to this precision, they serve better as building blocks of trading signals without statistical optimization and are easier to predict based on real-time information. Reason and evidence suggest that macro and market trends are complementary. Two combination methods are [1] market information enhancement of macro trends and [2] market influence adjustment of macro trends.

The below post is based on proprietary research of Macrosynergy.

Some links for data access and documentation refer to the J.P. Morgan Markets website and require user credentials. Yet, the links are not necessary for understanding the content of the post.

This post ties in with this site’s summary on information efficiency.

Market trends: some basics

“A trend is the general direction in which market values or the price of an asset evolve. Trends can be upwards, downwards or sideways…The longer the direction is sustained, the more qualified the trend becomes.” [Avatrade]

“Trend following has two natures. It is at one level a phenomenon of the human psyche…a burst of conformity to innovations in our immediate environment. At this level the masses have always been trend followers, not only in financial matters, but also in terms of music, art, clothing and basic world-views. The other level of trend following is…to design intelligent, premeditated responses to market action. This is the level of trend following from which we as traders…operate…Its essential elements…are [1] to initiate positions based on the perceived direction of the trend, [2] to hold positions based on the perceived direction of the trend, and [3] to liquidate positions based on the perceived direction of the trend.” [Ostgaard]

“The most common trend trading strategies use technical [market price-based] indicators. Popular trend trading indicators include moving average, relative strength index and average directional index…[In particular] a moving average indicator finds the average price of an asset over a given timeframe. By doing so, it creates a smoothing effect on the price data, producing a single line that can help traders identify trends. There are popular choices, such as the 50-day and 200-day moving averages, but ultimately the choice will depend on the individual…A common moving average strategy is to look for crossovers between two moving averages.”[IG]

Basic trend-following is cheap and simple to implement. All one needs is history and feed of market prices. What one does not need is a detailed understanding of contracts, valuation, or market and macroeconomic conditions. Moreover, trend following can principally be applied to all types of traded contracts. For pure abstract trend following it does not matter if one trades a stock, a crypto-currency, or a commodity future.

Macro trends: definition and importance

Macro trends here are defined as time series that characterize states of the broader economy that are plausibly relevant for market prices. For example, in countries with standard monetary policy regimes, inflation and operating rates are related to short-term interest rates and discount factors. Also, in most economies, nominal GDP growth and terms of trade are related to the outlook for corporate earnings.

Most macroeconomic data are public information and freely provided by government institutions. However, their interpretation is not always straightforward and usually subject to statistical adjustment, evaluation in economic models, and expert “commentary”. Moreover, standard publicly available economic data do not come in a format that is backtestable or tradable: publication time stamps are not provided, history has been compromised by revisions, and models have been applied with hindsight. In short, traders that want to incorporate macro trends in investment decisions must invest significant time and money to extract the information they need.

In the presence of information cost, markets will not track macroeconomic trends exhaustively and continuously. Indeed, an information efficient market produces, researches, and applies macro information only to the extent that investment returns and other benefits exceed information costs. Such rational inattentiveness reflects costs of acquiring information or costs of re-optimizing investment decisions. There is empirical evidence that inattentiveness causes sticky expectations and goes some way in explaining price momentum after important relevant news, such as corporate earnings releases (view post here). Oftentimes, herding is preferable for traders that cannot afford expensive research, particularly if some other market participants are likely to have better information, giving rise to “rational informational herding”. (view post here)

Competing market and macro trends

If quantifiable macro forces play a dominant role in a market, market trends and macro trends should be correlated. Hence, from a statistical angle, they can be seen as competing approaches to trend strategies. However, market and macro trends have different strengths and weaknesses:

  • Market trends are very timely in conveying information but provide little clarity in respect to the specific information that has been driving them. For example, bond yields may increase in response to either inflation or risk premia, two trends with similar symptoms in terms of price actions but different implication for subsequent expected returns.
  • Macro trends based on economic indicators are more specific in their information content but are often subject to long publication lags. For example, for a simple quarterly GDP growth rate (% over a year ago) the time elapsed between a first release and the mean of the observation period for which the change has been calculated is typically more than 200 trading days.

Here we apply two standard market and macro trends to international markets for short-term rates contracts, a market segment that is shaped predominantly by expectations for monetary policy and macro trends that influence central bank actions, such as economic growth and inflation. We are looking at 2-year interest rate swaps in the two largest currency areas (U.S. and euro area) and 13 smaller but still liquid markets: Australia (AUD), Canada (CAD), Switzerland (CHF), the UK (GBP), India (INR), Japan (JPY), South Korea (KRW), Mexico (MXN), Poland (PLN), Sweden (SEK), Thailand (THB), Taiwan (TWD), and South Africa (ZAR).

As a simple representative market trend, we use the classic 50-day versus 200-day moving average difference. These averaging parameters are standard insofar as they are most frequently given as examples in articles on trend following. Also, they are roughly consistent with the frequency domain of minor macroeconomic fluctuations.

As a simple representative macro trend, we use the average of excess inflation and excess economic growth. The indicators have been used in a previous post on macro trends (view post with details here) and accord with the intuition of the popular Taylor rule for monetary policy. In short, excess inflation is an average of various standard inflation rates minus the currency areas’ inflation target. Excess growth is an estimated GDP growth trend, based on recent high-frequency activity data, minus a 5-year moving median.

To be comparable with price data, the macro time series must correctly reflect the information state of the market and, hence, have been taken from the J.P. Morgan Macrosynergy Quantamental System (JPMaQS). These data measure the market information state of the latest instance of a macro-fundamental metric for any given day. Researchers and investors with access to J.P.Morgan Markets can gain access to the data and replicate the below analysis via API by following the instructions here.

The below charts show the simple sums of local macro trends since 2000 or as far back as good data were available. Note, however, that for strategy evaluation of the relevant macro trends for the smaller countries would usually be averaged with those of the large countries, since the large countries’ rates trends wield great influence over the small ones, but not vice versa.

Correlation between these simple market and macro growth trend metrics has been highly significant over the past 20 years on a panel basis for the large currency areas as well as for the set of smaller markets. Pearson correlation coefficients have been near 50% at a monthly or quarterly frequency. This confirms that trends in short-term rates markets are closely associated with the presumed key macro trends.

The close relation between market and macro trends suggests that they share common information trends, particularly those that reinforce shifts towards tighter or easier monetary policy. This does not necessarily mean that either of the trend metrics is satisfactory trading signals, but it does suggest that their explanatory power benefits from similar market forces, including rational information inattentiveness or disposition effects.

With such high correlation, it is not surprising that also the predictive power of market and macro trends have been similar over the past 20 years. In the two large currency areas, the correlation of market trends with subsequent monthly IRS returns has been near 12%, while the correlation of macro trends with subsequent returns has been over 14%.

Also, accuracy, the ratio of correctly predicted monthly market direction, has been higher for macro trends at 55.3% than for market trends (53.5%). Macro trends predicted both positive and negative market direction with a higher success rate. Finally, simple naïve PnL simulations, based on monthly repositioning in accordance with a trend z-score (winsorized at maximum 2) show a Sharpe ratio of 0.9 for macro trends versus 0.6 for market trends, not considering transaction costs.

For the 13 smaller markets the picture is a little different. First, correlation with subsequent IRS returns is generally lower, plausibly reflecting the more prominent role of interferences from exchange rates and capital flows. Also, return correlation is a little higher for standard market trends than for macro trends. This probably reflects the lower quality of growth and inflation data in some smaller countries.

Also, signal accuracy with respect to the subsequent monthly market direction in small countries has been a trifle higher for market trends (53.3%) than for macro trends (53%). And a simple naïve PnL, based on monthly trend z-scores would have produced slightly a higher Sharpe ratio for market trends (0.8) than for macro trends (0.7).

With simple market trends producing similar overall trading value while being much cheaper to produce the critical question is why we would ever need macro trends. The answer is that macro trend signals offer added features for performance enhancement, complementing the strengths of market trends:

  • Logical enhancements: Since macro trends represent specific information trends it is easier to combine several of them in a logically consistent manner. For example, information on growth or inflation can just be joined by weighted averaging. Supplementary information on monetary policy trends, such as credit growth of effective currency values may require some rescaling but can eventually also be brought to bear through simple addition. Qualifying information such as working day or holiday effects can be incorporated through standard adjustment procedures. Put simply, macro trends can be enhanced without the need for much statistical learning from predictive performance and, hence, can be operated with far greater complexity than market trends.
  • Logical breaks: Macro trends are easily interpretable and, therefore, more predictable. In particular, it is much easier to assess if real-time news implies a break in specific macro trends than in market trends. For example, news of natural disasters, pandemics, or credit crises implies unambiguous negative shocks for growth but not for equity and rates market trends, since the latter also need to weigh the consequences for inflation, monetary policy, and other market factors.
  • Complementary information: In the trade-off between interpretability and timeliness market and macro trends are on opposite ends of the spectrum and, therefore, can compensate for each other’s weaknesses. A strong case can be made for macro trends and market trends to be intelligently combined, with a minimum of domain knowledge. The below sections propose two such methods: market information enhancement and market influence adjustment of macro trends.

Complementary market and macro trends

Market information enhancement of macro trends.

The idea is simple: a macro trend measure is penalized, i.e. reduced, if and to the extent that the market trend subsequent to the period for which it is measured suggests a break in that trend. The market trend here is derived from the predicted contract’s returns. A break is suggested if the return trend runs opposite the direction that would normally be consistent with the macro trend. For example, a strong inflation and growth trend would be consistent with rising interest rates and negative fixed receiver swap returns. Thus, if the latest observation period of the economy suggested strong inflation and growth and the subsequent rates market trend implied declining yields, a continuation of the previous macro trend would look less likely.

We can implement this idea by using the information on the end-of-observation period and median observation period lags provided by JPMaQS for all quantamental indicators. These give for every real-time date the number of days that have elapsed since the end and the median of the observed period of the economic statistic respectively. Based on this information we calculate the difference between the latest value of the return index (5-day average) and its moving average over the macro trend’s observation period. Put simply, this proxies the market information that has become available since the latest economic trend has taken place. This is information that cannot possibly be in the macro trend numbers. Note that inflation and growth do not regularly have the same observation period and in this case, we average the relevant subsequent market trends.

Then we multiply the macro trend by the following modifying factor:

modifier = 1 + abs(mzc) x dir

where

  • mzc is the “clipped” market trend z-score based on panel standard deviation (winsorized at -1, 1), and
  • dir is the directional dummy, which takes 1 if fundamental macro trends and timelier market trends imply the same direction of target returns and -1 if they imply opposite directions.

This modifier has values between 0 and 2. If there is no market trend, the fundamental macro trend is 1, i.e., the macro trend is not modified. If the market trend is opposite in effect to the fundamental trend the modifier will be below unity up to a minimum of zero, which is attained if the market trend is 1 standard deviation in absolute size. If the market trend is equal in effect to the fundamental trend the modifier is above unity up to a maximum of 2, which is reached when the market trend reaches 1 standard deviation in absolute size. Put simply, depending on market and macro trend consistency, the macro trend signal can at best be doubled or at worst be set to zero.

The below shows the panel of macro trends modified by the subsequent IRS market trend in this way. Modifications are modest most of the time but can make a significant difference in episodes of large macro trend changes.

In the two large countries market information-enhanced macro trends have been slightly better predictors of IRS return trends than either market or macro trends individually. The correlation with subsequent monthly returns in the G2 has been over 15% (up slightly from 14.4%). Accuracy with respect to predicting monthly returns also was a bit higher at 55.9% (versus 55.3%).

For small countries, the adjustment would also have produced just a marginal increase in correlation (to 8.8% from 8.6%) and basically no change in accuracy (53%).

Market influence adjustment of macro trends

Market influence adjustment here means adjustment for economically relevant market trends subsequent to the observation period for which the latest macro trend has been recorded. For this adjustment one can use any relevant market, not just the one that has to be predicted. It is widely accepted that market moves themselves have effects on macro trends. For example, all else equal, widening credit spreads reduce economic growth, and appreciating currencies reduce local inflation in open economies.

In the present case, we postulate that declines in equity markets values indicate a reduced probability of monetary tightening, the very trend that is to be predicted by the inflation and growth macro trend. That is because equity price drops typically indicate a deteriorating growth outlook or financial tightening.

The adjustment is implemented analogously to the information enhancement above. The basis for the adjustment is the local equity index future return index. The relevant market trend is the difference between the latest return index value (5-day moving average) and the average return index value during the observation periods of estimated growth and inflation. The modification factor of the original macro return trend is calculated according to the formula in the previous section except that now the directional dummy is 1 when the macro trend is positive and the subsequent equity return trend negative. Hence, the positive macro trend signal for the rates market can double when equity markets post a significant drawdown and go to zero if they show large gains.

Looking at the panel time series of the adjusted macro trends one can notice that differences between adjusted and unadjusted series are more conspicuous than in the case of information enhancement. Some episodes of macro trends have been taken out and some greatly magnified.

In the two large currency areas, correlation of equity-trend adjusted macro trends with subsequent market trends has been notably higher than that of simple macro trends (18% versus 15%). Accuracy of the adjusted trend with respect to predicting subsequent monthly IRS returns reached 56.2, up from 55.3%.

Also, for the small countries equity trend adjustment would noticeably increase the predictive power of macro trends. Correlation with subsequent monthly returns would increase to over 13% from less than 9%, while accuracy with respect to the direction of subsequent monthly returns would exceed 54% versus 53% without adjustment.

 

Editor
Editorhttps://research.macrosynergy.com
Ralph Sueppel is managing director for research and trading strategies at Macrosynergy. He has worked in economics and finance since the early 1990s for investment banks, the European Central Bank, and leading hedge funds.