I. What is DeepMacro?
DeepMacro is based on two beliefs.
DeepMacro provides investors and policymakers with an automated, real-time view of the global macroeconomy, using novel "Big Data" sources, and applies these insights to markets.
The core of this system is based on dynamic factor models for medium-term growth and inflation developed at Platinum Grove Asset Management (PGAM). DeepMacro incorporates Big Data into this framework, to give early, orthogonal, and high-frequency insights into global economic trends.
Although automated, the framework is not a black box. It is consistent with a classic view of the business cycle as the co-movement of the activity of firms, individuals, and public entities, across sectors, geographies, and occupations.
The coverage is broad -- the G10 and major emerging markets. The system is “point-in-time”, meaning that all data are preserved as of the time they were released, which is what the markets had in the past when they were trading the information. And it is battle-tested, as it was constructed a decade ago at PGAM (where it was used to trade a systematic FX portfolio).
DeepMacro incorporates “unstructured data” from the Internet into this global macro framework. The global economy is driven by – perhaps defined by – information about all of humanity’s business activities and interconnections. On the Internet, billions of actors constantly radiate data about what they are doing, and about what will happen in the future. Like railroads and telegraphs in the 19th century or highways and the electricity grid in the 20th, the Internet can tell us a lot about the economy: what parts of the economy are expanding, what parts are being left behind, and how different parts are interacting.
DeepMacro captures several foundational sources of unstructured data from the Internet that can inform us about many types of economic activity across countries. We employ a variety of techniques to automatically extract and process these data, and incorporate them into the systematic framework for a fuller picture of the global macroeconomy.
Effectively, DeepMacro mimics the way that markets process information. Markets price both public information (official economic releases that are available to all) and private information that individual participants acquire on their own, such as anecdotes about what is happening in important economic sectors, their direct experience as consumers, and, increasingly, observation of supply and demand in certain products and services on the Internet. Our Big Data analysis attempts to do in a systematic fashion what the market does in a decentralized fashion. Each Big Data source is a piece of the private information that is available to certain actors, and aggregated up and reflected in market prices.
II. What does DeepMacro produce?
DeepMacro has three primary types of indicators:
Growth and inflation factors. These factors summarize information on the two most important macro variables for markets: growth and inflation. They are based on dynamic factor modeling to extract the level of business cycle and inflation activity relative to their long-term trends, and how quickly this activity is changing. The level of activity and the rate of change are the two dimensions of fundamental economic variables that influence how different asset classes perform through the cycle.
The factors are monthly time series updated about once an hour to capture all relevant data. The factors are designed to distill the trend from the noise. A single country’s business or inflation cycle will not change so often, but the relative order of countries does change, and it is the changes in the relative position of countries that create investment opportunities.
The factors are medium-term in nature. A large number of variables are filtered for leading relationships with broad measures of economic activity or inflation, and distilled into a common component that represents how the set of variables is moving relative to trend. Hence the growth factor is a way to operationalize one of the classic conceptions of the business: the co-movement and interaction of different parts of the economy (the same goes for the inflation factor).
The growth factor is not the same as GDP. The DeepMacro growth factor also incorporates a broader variety of data, such as employment, new orders, and, crucially, expectations for economic behavior. As such, the growth factor capture the living, breathing economy better than an accounting measure such as GDP does. At the same time, the growth factor is correlated with GDP, and regressions of GDP growth on the growth factor and other variables (such as lagged GDP) show that other explanatory variables do not add anything to the growth factors.
Daily Industrial Indicator (DII). The DII is an indicator of industrial activity that DeepMacro produces on a daily basis, for every country in the world. It is based on data on emissions that correlate with industrial activity: the production and transportation of goods, the generation of energy, and the like. Data are collected from the Internet for small areas (squares of about the distance from about the Empire State Building to Seton Hall in Newark NJ, so approximately 1 million sites globally), mapped into larger geographical areas, ultimately into countries, and filtered by strength of relationship with industrial activity.
“Precasts”. A precast is a forecast of an economic indicator based on unstructured data collected from the Internet. Because we collect our own original data on a continuous basis – i.e., well before the time series on which most Street economists base their forecasts are available – we call forecasts based on these data “precasts”. Macro Precasts help investors manage risks around events on the calendar, and help detect when new trends emerge. We currently produce precasts for the China and other countries’ manufacturing “purchasing managers’ indexes”, and US nonfarm payrolls.
DeepMacro also collects unstructured data on important economic phenomena that are not released in a monthly indicator yet are important to markets. For example, we examine social media and Internet infrastructure data to track sentiment on the Italian constitutional reform referendum. We also follow capital flows. Movements of capital across borders leave footprints in a variety of places on the Internet, and can meaningfully affect foreign exchange rates and asset prices.
Many indicators, integrated view. DeepMacro provides an integrated view of the key macro risks that affect asset prices: the medium-term business and inflation cycle, turning points in the cycle, and the event risks that even the best-constructed macro portfolio is exposed to.
III. What is the “growth factor”?
The DeepMacro growth factor is our measure of the business cycle. It is constructed in the same way across all countries, so we can compare countries on an apples-to-apples basis, even though they release different types of economic indicators, of different frequencies and at different times.
The growth factor
What is the business cycle? The dynamics and interactions of many economic phenomena. The growth factor is constructed from thousands of macro data series, which are selected and categorized in a consistent way across a large number of countries. These data series cover different parts of economic activity: spending (such as retail sales), sentiment (such as PMIs), the number of people (such as nonfarm payrolls), and others (such as home buyer traffic). State-of-the-art econometric methods (dynamic factor modeling, Stock and Watson 2002) are used to combine the data in a statistically optimal fashion, to construct a high-quality indicator of economic growth in each country. The growth factors are updated in real time by weighing the flow of new data releases, and incorporating the data into a simple estimate of the vitality of economic activities.
The growth factor is a monthly frequency series, and it updates hourly.
Coverage
The growth factor covers the G10 countries (including separate indicators for the major economies of the EUR area), plus major emerging markets. Coverage is being expanded to more emerging markets. The growth factor is computed for these countries about once per hour, so the DeepMacro system produces a complete picture of growth across our entire set of countries about 24 times per day.
Data
DeepMacro automatically collects and processes thousands of real and nominal economic time series in the growth factors. All data are seasonally adjusted, and converted into real terms in order to extract the “real” (as opposed to nominal) growth factor. New data series with short histories can be incorporated quickly into the system.
DeepMacro is incorporating “unstructured” data from its Big Data collection and analysis efforts into the growth factors. These data sources are high frequency, have little or no lag, and are not widely used by economists, so the incorporation of these data should give earlier, orthogonal views on the business cycle.
Point in time
The data are stored and analyzed “point in time”. Macro data are often revised, so the past looks different from today’s vantage point than it did in the past, when markets were processing and trading on the originally released information. Strategies based on this “forward looking bias” can often be erroneous, because revisions can be substantial, and markets did not have the revisions at the time of trading.
DeepMacro preserves all data as they were at the time of release, and all strategies are based on point in time data and methods.
Relationship with GDP
The DeepMacro growth factor is similar to GDP, and we can show forecasting performance of the growth factor for GDP for all our countries on request. However, the growth factor is different from GDP in some important ways. GDP is an accounting measure of money spent, earned, or produced. As such, it does not include sentiment, it does not include forward-looking expectations, and it does not include other variables that are important independent contributors to market perceptions of economic growth. For example, the unemployment rate and labor force participation are nowhere in GDP, but they are in the growth factor, and they are important to markets.
How we express the growth factor
From a market perspective, the two most important dimensions of the business cycle (or GDP) are its growth rate, and the change of change of this growth rate. DeepMacro constructs the growth factor to clearly show these dimensions. The growth factor itself is a measure of the rate of change of the underlying variables (versus its long-term average, which is set to zero). The momentum of the growth factor is simply the change over a desired horizon: one month, three months, etc.
How to use the growth factors
The growth factor can be used in many ways. It can be used as a systematic input into a discretionary investment process. By rapidly aggregating large amounts of information into a view for a country, that can be compared to other countries on a like-for-like basis, the growth factor is well designed for use in strategies that compare countries, such as foreign exchange, or cross-country equity strategies. The growth factors can also be aggregated into regional (such as “Asia”), or thematic (such as “commodity countries”), indicators, for use in theme-based macro investing.
The growth factors can also be used as building blocks of systematic strategies. At Platinum Grove Asset Management, the growth factors were used as a pillar in a systematic foreign exchange trading strategy. DeepMacro is releasing such strategies soon.