The aim of this workshop is to give delegates a practical understanding of the main statistical forecasting tools that are available to support marketing, production, supply chain, and financial decision making.
Who will benefit?
A background in statistics is not necessary to enjoy the course. You will need to be comfortable working with numbers, with evidence-based thinking, and tolerant of a little algebraic notation. Methods are however explained at an intuitive level, and delegates throughout get hands-on experience using the popular and accessible professional-standard forecasting software package EVIEWS.
Why is this important?
Every rational decision depends on a forecast. It is important for decision-makers to understand what techniques are available to make the best possible forecasts.
Choice of forecasting method depends on context, and the availability of information. In business, for long established products there is often plenty of data, and reasons to believe that sales may be readily predictable, at least in the short term. In financial markets, there is an abundance of data, but reasons to believe most price changes are close to random, making forecasting a real challenge, though changes in risk – volatility – are to some extent predictable
EVIEWS gives you easy access to all the methods relevant to these situations.
And everywhere – in business, finance, politics, demographics, crime, fashion, epidemiology, meteorology – enhanced computing power and behaviours revealed through social media have propelled us into an era of “big data”, which is being mined for regularities by a new generation of algorithms.
The methods explained and implemented in this workshop provide the foundation for these more complex “predictive analytics” methods. And in many cases, these simpler models give us more reliable forecasts.
What happens in the workshop?
Every session is a mix of taught principles, explaining how each method might be applied (on average 30% of time), and practical implementation of these methods using EVIEWS (70% of your time).
Delegates are encouraged to bring along their problems and data , and time will be set aside in the final sessions to discuss how EVIEWS can help improve their forecasting processes.
Forecasters face many different types of problem: – prediction of sales figures, inventory, earnings; risk and return on projects and investments; and events such as payment defaults and ratings changes.
Forecasters have at their disposal a rich array of statistical forecasting methods: regression methods (explanatory statistical models), and extrapolative time series methods (exponential smoothing, Box-Jenkins ARIMA models).
This session reviews evidence on best-practice in matching methods to problems, and at sources of data and other information to support the forecasting activity.
EVIEWS makes straightforward for users the processes of entering data and implementing current statistical forecasting tools.
This session covers simple methods of data entry and storage, the manipulation of raw data, and the use of EVIEWS to generate charts and summary statistics describing the forecaster’s data.
Linear Regression is a simple and popular method of forecasting, using prediction of one or more driver variables (say total consumer spending, product prices) to forecast another target variable (say, sales from an individual business).
This session shows users how to estimate the linkages between variables in this framework, how to test these for strength and significance, and how to use this information to produce predictions of the future path of the target variable.
We show how to describe seasonal fluctuations in data (dummy variables). How to identify the separate influences of driver variables when they often move together (collinearity). How to deal with situations where a driver variable such as price may itself be influenced by the target variable such as sales (instrumental variables estimation).
This session looks at practical problems often encountered by business forecasters using regression models.
Smoothing models were developed as simple methods to extrapolate a single time series of data, without looking at any causal factors. They break down the series into level, trend and seasonal components, each of which may vary over time.
This session introduces Single Exponential Smoothing, the Holt Linear model, and the Holt-Winters seasonal model, and shows how each can be used to make quite persuasive forecasts of univariate series even from a limited run of past data.
The Box-Jenkins ARIMA modelling process has been the workhorse of statistical time series analysis for many decades. It involves prior testing of single time series for “stationarity” (roughly, absence of trend); differencing of series to ensure stationarity; the identification of promising time series models in the ARIMA class.
This session uses the automatic forecasting capability of EVIEWS to carry out all these steps, and to make forecasts from single and combined ARIMA models.
To predict events – will a firm remain solvent or encounter financial distress? - we can use a regression model that has a 0-1 target variable. Incidentally, these models are the building blocks for the complex nonlinear “neural network” models often sued in big data mining.
This session shows users how to set up the required Logit Regression, and how to interpret the output from such models.
To predict the time-varying risks in, say, share prices or currency values, econometricians have developed regression-type models that predict not only the expected future value of, say, returns on a share portfolio, but also the risk surrounding this return.
This session shows users interested in financial market forecasting how to set up and interpret GARCH volatility forecasting models.
Often a forecaster will want to make forecasts of two or more variables that are simultaneously determined – sales volume and price, say: earnings, interest rates and exchange rates.
This session shows interested users how to set up multiple equation regression models, and multiple-equation pure time series models (vector autoregressions), and how to make forecasts from such systems.
Day 1 (Monday, 29 April 2019)
|Time||Session / Description|
|09:30 – 10:30||Session 1. What makes a good forecast? Review of Best Practice. Data Requirements and Sources|
|11:00-12:30||Session 2. Introduction to EVIEWS. Data Handling and Visualisation.|
|13:30 – 15:00||Session 3. Telling a Story: Regression Methods. Interpreting Results. Choosing between models.|
|15:30-17:00||Session 4. Practical Problems in Building Causal Models. Seasonality, Collinearity, Simultaneity.|
Day 2 (Tuesday, 30 April 2019)
|Time||Session / Description|
|09:30 – 10:30||Session 5. Keeping it Simple: Smoothing Models. Univariate Time Series Patterns. State Space Modelling|
|11:00-12:30||Session 6. Making it Complicated: ARIMA Models. Stationarity. Combining v Choosing.|
|13:30 – 15:00||Session 7. Advanced Topics in Forecasting. Predicting Events. Predicting Volatility.|
|15:30-17:00||Session 8. Forecasting Interdependent Variables. Explanatory Models. VAR Models.|