Model comparison with data sweeps

This slideshow requires JavaScript.

A useful exercise in modeling economic time series is to perform a “sliding window” analysis of the data that computes models in subsets of  the data and tests for the robustness of signal extractions, forecasts, and parameter variance relative to a growing subset of the data. For instance, for a time series of length 300, one could estimate a model on a shorter subset of the data, say for the first 200 observations, and then increase the amount of observations, re-estimate, and then see how the model parameter values change as the number of observations or data subset increases. One can also see how the signal extractions and forecasts change with additional data. Ideally, if the model is specified correctly for the data, there should be a very small variance in the estimated parameters as more data is added to the time series. It signifies the stability of the model selection. Normally, such an exercise would be tedious to carry out with X-13ARIMA-SEATS, or any other software such as MATLAB or R as scripts or spec files would have to be written for each individual re-estimation and then re-plotted. In the uSimX13 module of iMetrica however, this task has been rendered an easy one with the addition of a sliding windows tool.  In this blog entry, we describe this so-called “sliding windows” process and show just how fast and seamless it is to perform model choice robustness and comparisons in iMetrica.

We begin by describing the sliding span/window tool in the iMetrica-uSimX13 module. Once time series data has been loaded into the uSimX13 module from either the uSimX13 main menu or imported from the Data Control module, the uSimX13 computation engine must first be turned on from the uSimX13 menu. Then to access the sliding windows interface,  simply click on the “Sliding Span/Window Activate” check box in the main uSimX13 menu (see Figure 1).

Figure 1. Main drop down menu for the uSimX13 module, showing the “Sliding Span/Window Activate” check box.

Once clicked, the entire plotting canvas will turn to a dark shade of blue, which indicates the windowed region in which model estimation occurs. To control the sliding window, place the mouse cursor along one of the edges of the canvas and slowly glide the mouse with the left-mouse button held down either left or right, depending on which edge of the plot canvas you are on. Moving to the left or right with the left mouse button held down, the windowed area will shrink or expand. The model parameters are estimated instantaneously as the window adjusts and in effect, all the available model statistics, diagnostics, signals, and forecasts are computed as well. For example, as the window expands or shrinks, the trend, seasonally adjusted data, and 24-step ahead forecasts can be plotted and viewed in real-time as the window changes (see Figure 2). One can also slide the window to the left or right by placing the mouse anywhere inside the blue-windowed region, holding down the left mouse button and moving along the time domain. This way, the window length will remain fixed, but the window center will move along different subsets of the data. This can be useful for seeing how model parameters can change within regions of data that exhibit regime changes, namely a sequence in the series that suddenly changes in seasonal or cyclical structure after a certain time observation. The data can now be modeled in both sections before and after the regime change occurs in order to compare the estimated parameter values.

Figure 2. The window sliding across different subsets of the data. The signal extractions, forecast, and model parameters are recomputed automatically as the window changes. Forecast comparisons with the real data as the window span moves is now trivial. Here, the plot in cyan represents the original time series data in-sample and the 24 step forecast out-of-sample, and the light green plot is the time series data adjusted for outliers, as indicated in the model box. One can select the plots using the “series components” plot box. The data in gray represents the time series data not used in the model estimation.

Data Sweep

With the ability to seamlessly capture partitions of the data and model within the given partition using the sliding window, a natural extension of this mouse-on-canvas utility is to employ it somehow in comparing different models of the time series data. We call this method of model comparison time series data sweeping (or simply data sweeping) and it involves selecting an initial window of data from the first observation to the n-th observation where n is some number much less than the total number of observations $latex N$ in the data set (say, one third the amount). The data sweep then computes the sliding window from n as the final observation all the way to N, in increments of one (see Figure 3). At each addition to the length of the window, the forecast is computed for up to 24 steps ahead. Of course, since the true time series data is known in the out-of-sample region of computation, we can compute the forecast error for up to h \leq 24 steps ahead and sum up these errors as n increases to N. We can do this data sweep for several models, computing the aggregate forecast errors over time. The idea is that the best model for the data will ideally have the smallest forecast error, and thus comparing this forecast error with several models will identify the model with the best overall forecasting ability.

To access the data sweep, simply go to the main uSimX13 menu, shown in Figure 1, and click “Sweep Time Series Control Panel”. This will bring up the main interface for the data sweep (shown in Figures 4-6). To begin the sweep, first select the model and regressors desired to model the data with inside the model selection panel of the main uSimX13 interface. Then choose at which observation you’d like to carry out the data sweep (starting at observation n=60 is the default). Lastly, select how many forecast steps you’d like to use in computing the forecast error (1-24). Once content with the settings, click the “Compute time series sweep” button and watch as the window span increases from n to N, recomputing parameters, signals, and forecasts at each step (see slideshow at top of post).  Once the sweep is complete, the parameter statistics, Ljung-Box mean value at two different lags, and the total forecast error is displayed in the control panel. To compare this with another model, save the results of the sweep by clicking “save parameters” in the uSimX13 menu, and then choose another model and recompute (while using the same settings as the previous sweep, of course).

To give an example of this process, we begin by simulating a time series data set of length N = 300 from a SARIMA model of dimension (0,1,2)(0,1,1)_{12}, namely a seasonal auto-regressive integrated moving-average process with two non-seasonal moving-average parameters, and one seasonal moving average parameter. The data sweep is performed on the simulated data with a forecast error horizon of length 23 using three different SARIMA models, (a) (0,1,1)(0,1,1)_{12}, (b) (1,1,0)(0,1,1)_{12}, (c) (0,1,2)(0,1,1)_{12}, the true model. See Figures 4-6 below to see the data sweep results and the estimated parameter mean and standard deviation, the average Ljung-Box statistics at lag 12 and 0, and the forecast errors for each model. Notice the forecast error for the true model (c) (figure 6) is the lowest followed by model (b) (figure 6) and then (a) (figure 4), which is exactly what we would want.

Figure 4. Model (a) and the parameter statistics, forecast error, and data sweep controls.

Figure 5. Model (b) and the parameter statistics, forecast error, and data sweep controls.

Figure 6. Model (c) (the true model) and the parameter statistics, forecast error, and data sweep controls.

Advertisements

iMetrica and Hybridometrics: Introduction

The high-frequency Financial Trading interface of iMetrica. Easily construct in-sample trading strategies with an array of optimizers unique to iMetrica and then employ the strategies out-of-sample to test and fine-tune the trading performance.

This blog serves as an introduction and tutorial to Hybridometrics using iMetrica. Hybridometrics is a term used to express the analysis, modeling, signal extraction, and forecasting of univariate and multivariate financial and economic time series data using a combination of model-based and non-model-based methodologies. Ideal combinations of computational paradigms and methodologies used in hybridometrics include, but are not limited to, traditional stochastic models such as (S)ARIMA models, GARCH models, and multivariate stochastic volatiluty models   combined with empirical mode decomposition techniques and the multivariate direct filter approach (MDFA). The goal of hybridometric modeling is to obtain signal extractions and forecasts, for official use or government use, all the way to building high-frequency financial trading strategies, that perform better than using only model or non-model based methods alone. In other words, hybridometrics seeks to extract the advantages of different paradigms combined to outperform traditional approaches to time series modeling. The iMetrica software package offers the most versatile and computationally efficient portal to this newly proposed time series modeling paradigm, all while remaining surprisingly easy to use.

The iMetrica software package is a unique system of econometric and financial trading tools that focuses on speed, user interaction, visualization tools, and point-and-click simplicity in building models for time series data of all types. Written entirely in GNU C and Fortran with a rich interactive interface written in Java, the iMetrica software offers an abundance of econometric tools for signal extraction and forecasting in multivariate time series that are both easily accessible with the click of a mouse button and fast with results computed and plotted instantaneously without the need for creating output data files or calling exterior plotting devices.

One powerful feature that is unique to the iMetrica software is the innate capability of easily combining both model-based and non-model based methodologies for designing data forecasts, signal extraction filters, or high-frequency financial trading strategies. Furthermore, the strategies can be computed and tested both in-sample and out-of-sample using an easy to use built-in data partitioner that effectively partitions the data into an in-sample storage where models and filters are computed and then an out-of-sample storage where new data is applied to the in-sample strategy to test for robustness, over-fitting, and many other desired properties. This gives the user complete liberty in creating a fast and efficient test-bed for implementing signal extractions, forecasting regimes, or financial trading strategies.

The iMetrica software environment includes five interacting time series analysis modules for building hybrid forecasts, signal extractions, and trading strategies.

  • uSimX13 – A computational environment for univariate seasonal auto-regressive integrated moving-average (SARIMA) modeling and simulation using X-13ARIMA-SEATS. Features an interactive approach to modeling seasonal economic time series with SARIMA models and automatic outlier detection, trading day, and holiday regressor effects. Also includes a suite of model comparison tools using both modern and goodness-of-fit signal extraction diagnostics.
  • BayesCronos – An interactive  time series module for signal extraction and forecasting of multivariate economic and financial time series focusing on Bayesian computation and simulation. This module includes a multitude of models including ARIMA, GARCH, EGARCH, Stochastic Volatility, Multivariate Factor Stochastic Volatility, Dynamic Factor, and Multivariate High-Frequency-Based Volatility (HEAVY), with more models continuously being added. For most of the models featured, one can compute the Bayesian and/or the Quasi-Maximum-Likelihood estimated model fits using either a Metropolis-Hastings Monte Carlo Markov Chain approach (Bayesian) or a QMLE formulation for computing the model parameters estimates. Using a convenient model selection panel interface, complete access to model-type, model parameter dimensions, prior distribution parameters is seamlessly available. In the case of Bayesian estimation, one has complete control over the prior distributions of the model parameters and offers interactive visualization of the Monte Carlo Markov Chain parameter samples. For each model, up to 10 sample 36-steps ahead forecasts can be produced and visualized instantaneously along with other important model features such as model residuals, computed volatility, forecasted volatility, factor models, and more. The results can then be easily exported to other modules in iMetrica for additional filtering and/or modeling.
  • MDFA – An interactive interface to the most comprehensive multivariate real-time direct filter analysis and computation environment in the world. Build real-time filters using both I-MDFA and Zero-Pole Combination (ZPC) filter constructions. The module includes interactive access to timeliness, smoothing, and accuracy controls for filter customization along with parameters for filter regularization to control overfitting. More advanced features include an interface for building adaptive filters, and many controls for filter optimization, customization, data forecasting, and target filter construction.
  • State Space Modeling – A module for building observed component ARIMA and regression models for univariate economic time series. Similar to the uSimX13 module, the State Space Modeling environment focuses on modeling and forecasting economic time series data, but with much more generality than SARIMA models. An aggregation of observed stochastic components in the form of ARIMA models are stipulated for the time series data (for example trend + seasonal + irregular) and then regression components to model outliers, holiday, and trading day effects are added to the stochastic components giving ultimate flexibility in model building. The module uses regCMPNT, a suite of Fortran code written at the US Census Bureau, for the maximum likelihood and Kalman filter computational routines.
  • EMD – The EMD module offers a time-frequency decomposition environment for the time-frequency analysis of time series data.  The module offers both the original empirical mode decomposition technique of Huan et al. using cubic splines, along with an adaptive approach using reproducing kernels and direct-filtering. This empirical decomposition technique decomposes nonlinear and nonstationary time series into amplitude modulated and frequency modulated (AM-FM) components and then computes the intrinsic phase and instantaneous frequency components from the FM components. All plots of the components as well as the time-frequency heat maps are generated instantaneously.

Along with these modules, there is also a data control module that handles all aspects of time series data input and export. Within this main data control hub, one can import multivariate time series data from a multitude of file formats, as well as download financial time series data directly from Yahoo! finance or another source such as Reuters for higher-frequency financial data.  Once the data is loaded, the data can be normalized, scaled, demeaned, and/or log-transformed with a simple slider and button controls, with the effects being plotted on the graphic canvas instantaneously.

Another great feature of the iMetrica software is the ability to learn more about time series modeling through the using of data simulators. The data control module includes an array of data simulating panels for simulating data from a multitude of both univariate and multivariate time series models.  With access to control the number of observations, the random seed for the innovation process, the innovation process distribution, and the model parameters, simulated data can be constructed for any type of economic or financial time time series imaginable. The different types of models include (S)ARIMA models, GARCH models, correlated cycle models, trend models, multivariate factor stochastic volatility models, and HEAVY models. From simulating data and toggling the parameters, one can visualize instantly the effects of the each parameter on the simulated data. The data can then be exported to any of the modules for practicing and honing one’s skills in hybrid modeling, signal extraction, and forecasting.

Keep visiting this blog frequently for continuous updates, tutorials, and proposals in the field of econometrics, signal extraction, forecasting, and high-frequency financial trading. using hybridometrics and iMetrica.