Nowcasting, Forecasting the Now

Data powers the world, and it powers economic decisions. Government agencies, private institutions, and even the average Joe investor down the street all collect and monitor data to scope out the economy. However, this data is numerous and covers multiple facets of the economy, making it difficult to arrive at a single, instantaneous conclusion about the state of the economy. That's before nowcasting came in the picture.

Here’s an intro to nowcasting—complete with a background on economic modeling, key nowcasting concepts to know, how it works, its practice in banks today, and a closing piece on its significance.

A Background on Economic Modeling

Power is at the fingertips of those who consistently and accurately predict the economy. For investors or economists, knowing what the economy will look like in the next phase is like knowing a winning Powerball ticket’s coordinates before being called. With relevant economic information, businesses could get a headstart on aligning their capital decisions with upcoming economic trends; investors could pick different sectors of stocks that perform best depending on the predicted economic phase.

Many go to great lengths to build an economic model that will predict the economy. The linked paper mentions how almost all economic models incorporate variables like “economic growth, population growth, the labor market, production, consumption, savings, the capital market, inflation, productivity, technology, etc.” Of course, people have unique economic modeling styles and differ in what variables they choose or the weight they place on each.

Current Issues with Economic Modeling

Economic lag with certain variables. GDP, or Gross Domestic Product, is a key indicator of how the economy is doing. If it goes up, it means the economy is doing well; if it goes down, then not so much. Despite its popularity as a telltale symbol of the economy, GDP is a lagging indicator or is always a bit behind reality.

The curse of dimensionality. Economic models that are too simple lack the depth needed to sufficiently carry out predictions, while overcomplicated models may create models that become sensitive to variables that don’t matter or speak to the bigger picture. Finding the sweet spot requires a deep understanding of the forces affecting the economy.

Big, unmanageable data. To target the issue of economic complexity, Economists Burns and Mitchell researched the role data selection plays in building models, as well as the problem of big data: “data sets of much larger size, higher frequency, and often more personalized information” (wol.iza.org). In their work, Burns and Mitchell uncovered the “systematic co-movement among the series and pervasiveness of fluctuations across different sectors and different kinds of economic activities.” In simple words, different variables and sectors correlate in certain ways, as they are all relevant forces undulating the economy. Big data could provide better predictions of economic phenomena but is taxing to incorporate it into models.

What is Nowcasting?

Nowcasting literally means to forecast the “now,” or present. Nowcasting is a cool, modern approach to economic forecasting that applies machine learning. It’s able to account for larger clusters of data and accurately predict certain variables. One objective of nowcasting is to aggregate as much data—especially alternative data—and put it into a model that will forecast growth and inflation.

Right now, nowcasting is all the rage in the markets, since it’s a newfound way to pull multiple sources of large data and integrate it into a conclusive economic prediction.

Nowcasting Economic Growth

As mentioned earlier, GDP is a lagging indicator of the well-being of the economy. Nowcasting can address this issue by generating an early insight into economic GDP by pulling from recently published data to update GDP in real-time (imf.org). Something to note, nowcasts for GDP are meant to be short-term and applied to the current quarter.

Nowcasting Inflation

Nowcasting may also be used to forecast the current level of inflation in a given month or quarter. Before the official CPI or PCE inflation data are released, inflation nowcasts can provide a sense of where inflation is now and where it likely will be in the near future. Because GDP is a lagging indicator, perhaps a better way to gauge the economy is to focus on ISM Manufacturing Index—a leading indicator for inflation (clevelandfed.org).

Key Nowcasting Terminology

Understanding nowcasting requires understanding a few related terms. Below are a couple of nowcasting-related concepts that make the rest of the read easy—especially for those without much prior exposure to financial and statistical modeling.

Factor Models. “Factor” is synonymous with “variable”; factor models consider multiple variables, or factors, to monitor macroeconomic conditions and arrive at a conclusion about the economic state. They rely on the basic assumption that information about different aspects and sectors of the economy can be considered measures of the economy as a whole. Factor models are popularly used to aid asset managers in their investment decisions, as they help compute an asset’s future price or return. They come in many different types—as listed on CFI.

Dynamic Factor Models. Dynamic factor models (DFMs) are a type of factor model that takes into account the time-varying nature of the underlying factors. In other words, they model the evolution of these factors over time to better capture the dynamics of the data. They are appealing and accurate because DFM’s fit the data, describe the comovements of many macroeconomic variables—whether its “employment, average workweek in manufacturing, initial claims for unemployment insurance, [or] stock prices” (bls.gov)—despite being a single index, and can efficiently incorporate latent factors into the mix (princeton.edu).

Latent Factors. Latent factors are factors that are not directly observable or quantifiable. Some examples of latent factors include things like quality of life, business confidence, and customer satisfaction—all of which are not easily or directly measurable.

Kalman Filtering. Kalman filtering is a statistical process that efficiently deals with any missing observations by providing calculated fill-ins for the missing data. It makes data collection efforts for the target nowcasting variable (GDP, ISM, etc.) more easily obtained.

How Does Nowcasting Work?

Coding and machine learning certainly enhances nowcasting by applying automated algorithms to forecasting. Those models that incorporate machine learning tend to produce nowcasts superior to typical models because they are able to handle high-dimensional information sets. That’s one part, while the other component of building nowcasting models is identifying the optimal variables, periods, and lag terms (imf.org).

Nowcasting is often powered by dynamic factor model technology and involves an advanced form of Kalman filtering—of the kind used in robotics. The technology equips nowcasts with the ability to effortlessly capture economic variables and the co-movement phenomenon by accommodating all relevant data series—those typically too large and complex to input into economic models. Nowcasting “exploits the fact that these data series, although numerous, co-move quite strongly so that their behavior can be captured by few factors” (research.macrosynergy.com). The technology helps automate the puzzling process of which factors are the most useful for forecasting the economic variable of choice—whether it’s GDP or ISM or CPI. After all, the trick behind a well-developed model is to know what information to retain and what to discard.

For brilliant technical and statistical explanations of the process of Nowcasting, check out this bona fide paper and this page on Science Direct.

Nowcasting Used by Banks Today

Various well-known banks—names like JP Morgan Chase, Federal Reserve Bank of New York, Federal Reserve Bank of Atlanta, Bank of England, and European Central Bank—view nowcasting as a neat aid to economic awareness.

JP Morgan Chase Bank

JP Morgan utilized a specific COVID-related nowcasting model to determine the risk of recovery and probability of economic expansion within three months. 150 different variables that were suggestive of both people and money movement went into the model—alternative variables like open data reservations, changes in Uber rides, etc. Around COVID, the model ended up being wrong despite the model’s sophistication.

Investment Banks and Blockchain

Blockchain is emerging in investment banking: ALM, trade reconciliation, clearing and settlement, deal escrow and collateral management.

Disha BhowmickDisha Bhowmick

Federal Reserve Bank of New York

The Federal Reserve Bank of New York (FRBNY) publishes its own nowcast of US GDP growth. FRBNY’s model uses Kalman filtering techniques and a dynamic factor approach to arrive at their unique nowcasting “design that digests the data as ‘news,’ mimicking the way markets work” (newyorkfed.org).

Federal Reserve Bank of Atlanta

The Federal Reserve Bank of Atlanta’s GDP Nowcast has become a really popular model that people in the market look at to anticipate what's going to happen with upcoming data. According to their website, the FRBA mentions how their methodology for estimating GDP growth is similar to the one used by the U.S. Bureau of Economic Analysis, and they mimic the expenditure approach to calculating GDP using dynamic factor modeling.

Bank of England

The Bank of England’s Monetary Policy Committee (MPC) uses a compilation of nowcasts from three different models to gather their initial view on the economy’s present state. Indeed, the MPC’s nowcasts inform their month-to-month monetary policy decisions. Their nowcasting model incorporates different industries such as retail services, manufacturing, construction, etc. (production approach for GDP), a mixed-data sampling model, and a dynamic factor model.

European Central Bank

The European Central Bank (ECB) also refers to nowcasting models to help inform key policy decisions. In fact, ECB staff have contributed to commendable nowcasting research and have published various papers regarding the subject—such as “Now-casting and the real-time data flow.”

Why is Nowcasting Important?—Conclusion

Economic forecasting may be a shot in the dark, but nowcasting can bring more accurate predictions to light. While building nowcasting models from scratch may be difficult to do as it requires coding, economic, and model-building expertise, monitoring nowcasting models are highly useful and sets forecasters one step ahead of the game.

Nowcasting is unmatched by any other technique in its ability to handle vast amounts of data and complexity. And, nowcasting is currently in practice by banks. The Atlanta Fed GDP Nowcast is often used as a sign of the next GDP print. That said, nowcasting may not always yield accurate results. Consider the example of JP Morgan’s, where the variable relationships assumed by the model did not hold true. This can be highly costly, as access to the data inputted in models costs money. Like any other model, nowcasting models hold flaws. It all comes down to the competency of the modeler.

Nowcasting is important because it helps us predict business cycles. After all, business cycles influence asset prices, stock prices, bond prices, credit spreads, performance, styles, and factors. Indeed, the cycle drives company profits, business plans, employment levels, and consumer purchasing power. And, business cycles influence how money is flowing via monetary policy decisions and overarching economic policies. Early economic insight is key to gaining a holistic conclusion about the economy—and nowcasting may unlock it all.