To monitor economic conditions in real-time, we should use a much more modern approach known as Nowcasting since economic dynamics drive policy decisions, financial flows, and corporate profits and account for a big part of the price fluctuation of assets. Nowcasting helps to make financial market trading a lot more efficient as well. The dynamic factor model is the leading technology that is behind Nowcasting. It condenses all the information of multiple correlated “soft” and “hard” data series into a smaller number of “latent” factors.
The factor that is the most correlated with a growth-related data series of a diverse representative set can be interpreted as a growth nowcast. Economic markets read economic data real-optimized by the dynamic factor’s state-space representation. Kalman filter is the related estimation used to projections for all the data series and each release of data, a model-based surprise known as “news.” Nowadays, machine learning models, like elastic net, support vector machines, feed-forward artificial neural networks, and LASSO – these have all been deployed to improve nowcast’s predictive power.
Nowcasting is what we call the prediction of the present, the very recent past, and the very near future. This term has been used for a long time now in meteorology and is an abbreviation for now and forecasting. You may be wondering how it’s relevant to economics then. Nowcasting is appropriate there since you can only get the present economic state with a significant delay. This is even more true for the ones that are collected quarterly.
It essentially defines the more modern approach to monitor the present economic conditions in real-time. While the main thing that weather forecasters need to do is predict future weather and know the current weather conditions, economists must forecast the present, the near future, and the recent past. The main idea of this is to interpret and analyze the flow of the microeconomic news by frequently updating the key variable predictions, such as the real GDP (gross domestic product) growth, for each of the data releases. It helps to make short-term estimates of target variables that are lagging.
Why is Nowcasting Important?
The microeconomic surprises explain a big part of the fluctuations in asset prices. In the quarter-to-quarter changes of government bonds, up to one-third of it yields. There can be no one indicator able to be the silver bullet capable of solving all the problems that come with accurately tracking the economy’s evolution in real-time. On the other hand, combining the information that is contained in various available releases is an approach that is more likely to be promising.
Unlike the professional forecasters who apply a sort of judgment after combining several unrelated models, you are allowed a more transparent analysis of the data flow in real-time with the help of a single format model. What this model essentially does is that it summarises within an econometric framework the best expert knowledge and practice in the analysis of business cycles.
One of the most critical drivers of asset returns is the current state of the economy. This cycle drives employment levels, business plans, consumer purchasing power, and company profits. It impacts economic policy by directing monetary policy decisions. GDP may be the official measure that covers the whole economy. However, it is not all that useful from the point of view of an investor.
First off, it is only quarterly available, and it is also published with a significant delay. The first estimate is usually being released four weeks after the reference quarter has ended, which is too late to make decisions in real-time. Second off, it is noisy and, it is heavily revised, often even after years of the first release, despite it being an official measure. Nowcasting can help fill in that gap for the investors and provide a complete signal about the underlying economic activity – that too in real-time. It has become a tool essential for real-time analysis based not on stories. Still, facts digest high-velocity asynchronous data that arrive at arbitrary rates to estimate target variables of low velocity by using statistical techniques. Nowcasting has remained unmatched by any other technique that can use vast amounts of data and handle complexity.
Facts digest high-velocity asynchronous data that arrive at arbitrary rates to estimate target variables of low velocity by using statistical techniques.Tweet
Kalman Filter, Dynamic Factor Model, State-Space
The model for Nowcasting uses a heterogeneous and extensive set of predictors that include both “soft” and “hard” data, such as everything from consumer surveys to unemployment statistics. Even though these data series are numerous, the estimation procedure has exploited the fact that they can co-move strongly, so their behavior can be captured using a few factors. All the output series are generated using a dynamic factor model. The reason for this type of model being chosen was so that we could deal with the “curse of dimensionality,” which is many series that are correlated since we are required only to estimate a limited set of parameters for an extensive dataset.
To optimally exploit the dynamic relationships among them, the model assigns weights to the series. We can interpret nowcast as a growth component that is highly correlated to all the data series that are input: it also disregards information that is idiosyncratic while capturing standard signals given out by all microeconomic data releases, including surveys.
The dynamic factor model is often the engine used in Nowcasting, and it’s equipped with some advanced filtering techniques often seen in robotics. These methods are pretty standard in big analytics since they can effectively summarize all large data sets using a few common factors. Due to the advances in the development of high dimensional data econometrics, the development of Nowcasting has been possible. Everything from inventories to the purchasing managers’ sentiment to manufacturing, from international trade and transportation services to labor market indicators, is covered by the data set.
This approach is based entirely on automated procedures. The news flow is processed naturally, just how it would be by any other informed person. The surprise component is first extracted from the data. These surprises are then translated into a standard unit. This is the impact they are giving on the key macroeconomic indicators, say, nowcast GDP or corporate profits. When the technique for analyzing a system that has multiple outputs and inputs, known as state-space representation in statistics, was introduced to the models, Nowcasting had a significant breakthrough. The state-space picture expresses the variables as vectors, meaning quantities that consist of a magnitude and direction. A vector of observed variables is presumed to be explained by several unobserved factors in this representation. These hidden or “latent” factors are seen to follow an autoregressive relationship, meaning that their current values are dependent on their past values.
We can formalize how the macroeconomic data releases are read by the market participants in real-time from the space-state frameworks. This also involves monitoring a lot of data, forming expectations on them. Whenever the realizations diverge noticeably from those expectations, it’s needed to revise the assessment on the economy’s state. This can be possible because, for a model that is in space state representation, the Kalman Filter will generate projections for all the considered variables and is then allowed to compute, for every data release, a model-based surprise, the news.
The Job of Machine Learning
It is seen that after training a variety of popular machine learning algorithms over an expanded window and replicating a nowcasting situation, the results show that most of the machine learning models are producing point nowcasts, which are better than the simple autoregression benchmark. Top-performing models like LASSO (Least Absolute Shrinkage and Selection Operator), neural networks, and support vector machines can reduce the usual and standard errors made by nowcast approximately by 16 to 18%. To give you further performance improvements, it even uses various weighting schemes to combine the nowcasts of the ML models. When it comes down to Nowcasting and forecasting, it is also seen that neural networks can perform better than dynamic factor models when it is implemented in Nowcasting and forecasting by comparing the GDP growth forecasts of the artificial neural network to that of state of the art and latest dynamic factor models.
Important Empirical Findings
Models of Nowcasting can provide a degree of similar accuracy to the official preliminary estimates of the GDP growth, even though the nowcasting estimates are timelier. It is also seen through some academic research that there is a significant improvement on the standard statistical forecasting models, which make use only of the past data being represented by the nowcasts.
Small data approaches are what came in first, and they generally involve estimates of maximum likelihood. In contrast, consequent significant data approaches are typically engaged with a two-step estimation based on a first step extraction of the principal components. The big data nowcasting is not necessarily any better. To begin with, they tend to be a lot more tedious to manage and less transparent. They might not even be able to deliver a lot of improvement in the accuracy of factor extraction. This stabilizes and increases quickly as the numbers of the indicators go up. It is worth noticing that distortions can be created in the extracted factor with a set of poorly balanced indicators whereas, on the other hand, with the help of small data approaches, we can facilitate and promote hard thinking about a group of indicators well-balanced.
In financial markets, and the world of the economy in general, Nowcasting has brought about a significant change that helps us forecast economic variables with a lot more precision and ease. We hope this article helped you in understanding the basics and importance of Nowcasting.