Use Case: a real time leading index for the United States
The Leading Index for the United States, published by the Federal Reserve bank of Philadelphia, tends, as the name implies, to lead recessions (illustrated by shaded areas right). As such it is a useful metric of economic activity. However, the Leading Index is typically published with a lag of a month or more.
For metrics such as the leading index to be useful for real time decision making, they must incorporate all available information to date. Regarding leading indicators for the United States, OttoQuant users can accomplish this in two ways. One approach is a real time nowcast of the Leading Index itself. A second approach is a synthetic index of the relevant input data, without trying to imitate the Leading Index.
Nowcasting the Leading Index for the United States
The first approach allows us to estimate the Leading Index in real time, using daily and weekly data as well as monthly series that are published early, such as purchasing managers indexes or manufacturing indexes. Some of these components, such as the spread between the 10-year Treasury bond and the 3-month Treasury bill, are in fact components of the Leading Index, so that our early estimates should be an accurate indication of what the Leading Index will be when it is finally published. For this example, we’ll accomplish this by selecting New Custom Project >> Predictive Modeling >> Bayesian Dynamic Factor Model, selecting, “United States, Leading Economic Index” as our target variable, then selecting the covariates used to predict our target.
Results (left) include true data (orange), our nowcast (blue), and an ARIMA estimate for comparison (green). Clearly, ARIMA models are not a very useful tool in times of crisis. In addition to these results, OttoQuant also provides “Data Flow” for each date. By clicking a date, users can see exactly how each series impacted our predictions for the Leading Index. In addition to results, users can download input data and, for technical users, parameter estimates. Users can also backtest models with a few simple clicks. Backtesting uses only data that was published by a given date to examine out of sample performance.
A Synthetic Index for the United States
A second approach to indexing real time activity in the United States is to create a synthetic index from a panel of time series data without restrictions on what exactly this index should look like. The result, based on OttoQuant’s statistical routines, will be the index with the maximum possible predictive power for the data in the model. We accomplish this in the OttoQuant interface by selecting New Custom Project >> Signal Extraction >> Bayesian Dynamic Factor Model, and selecting the inputs to the model.
The great advantage of an entirely synthetic index such as this is its emphasis on real time data, with a high incidence of daily and weekly indicators, and that it is unconstrained by trying to predict a specific series. Instead, we have allowed the data to inform us as best it can on the state of the economy in real time. In addition to the index itself, our interface again indicates the impact of each input series in the “Data Flow” section by simply clicking on a date. Note that in this example we have opted to generate a single index. One can, if desired, generate multiple contemporaneous indexes, allowing a more detailed view of the state of input data. Because these indexes are unconstrained by target variables, they will naturally capture volatility in different sectors of the economy, such as labor, bonds (spreads), or manufacturing.
Inputs to Synthetic Index Model
AAA – T bill spread
AAA – BAA spread
Lithium spot price
Forward Inflation Expectations 5 y
Inflation Expectations 10y
USD exchange rate
10Y T bill
10Y – 2Y T bill spread
10Y – 3M T bill spread
Banks Balance Sheet
30y – 15y Mortgage Spread
Continuing Jobless Claims
Initial Jobless Claims
NY Manufacturing Index
Philadelphia Fed Manufacturing Index
Richmond Fed Manufacturing Index
Dallas Fed Manufacturing Index