[MUSIC] Welcome to week one of Excel skills for business forecasting time series models. Let's get straight into it. There are a number of classifications of business forecasting methods. Qualitative versus quantitative, time series versus causal. These classifications are not mutually exclusive. The business forecaster needs to be aware of the appropriate method to match the forecast situation. In business A large number of forecast will be of variables that are measured quantitatively, sales, costs and exchange rates for example. The distinction between quantitative methods and qualitative methods is about how the prediction is derived. For quantitative methods, the prediction is derived using some algorithm or mathematical technique based on quantitative data. For qualitative methods, the prediction is based primarily on judgment or opinion. The distinction between time series methods and causal methods is about how many variables are utilized to produce the forecast. Time series methods rely on the past measurements of a single variable of interest and no other variables, which we will focus on in this first course on business forecasting. For causal methods, the prediction of the target series or variable is linked to other variables or time series, which we will focus on in our second course on business forecasting. Predictions and forecasts are based on relevant current and past data. The data sources can be classified into internal and external. Internal sources come from within the organization, for example, sales data, employment records or customer profiles. External data is sourced from outside the organization, for example, bureau statistics data, government agencies, or commercial data agencies. A useful classification of data for forecasting is whether the data is time series data or cross-sectional data. Time series data is a sequence of measurements on a single variable taken over specified successive intervals of time, for example, monthly interest rates, sales per week or tourist arrivals per month. Cross-sectional data are measurements on a variable that are at one point in time but spread across the population, for example, tourism spending across different cities or production across different businesses in the country. In this course, we will focus on time series forecasting methods using time series data. In our second course, we will focus on causal forecasting methods. So how does business forecasting work in a nutshell? Well, it's a three step process. One, an evaluation of the time series for historical patterns. Two, matching this observed pattern to irrelevant algorithm, the forecasting method. Three, projection of the algorithm into the future to produce our forecasts. Okay, let's talk about these patterns. There are various patterns that are typically associated with time series data. These patterns can usually be linked to various components of time series. The systematic components are typically due to explainable factors. The forecaster needs to understand the components of the time series to match the appropriate forecast method or algorithm. The components of a time series are level, trend, seasonal, cyclical and random. The random component is the only non-systematic component which cannot be forecasted. This means that the level, trend, seasonal and cyclical components are all systematic components which can be forecasted. Let's look at the level components. The level indicates the underlying value of the series on the vertical axis for a given time period. The level of the time series may be constant over one time period, and it may change to another level for a different time period due to the influence of other components. If the level remains relatively constant over the entire time series, a horizontal data pattern is observed. For the trend component, there is a tendency for the underlying level of the time series to systematically increase or systematically decrease from period to period. The trend need not be consistent over the entire time series and trends need not be linear. Trends are usually caused by population change and technological change. The seasonal component is a systematic and repeatable fluctuation in the time series that usually occurs within a well-defined time period, such as the year or a week that is, the periodicity is regular. These fluctuations typically repeat themselves in future iterations of the set time period. Seasonality occurs due to weather or institutional reasons, for example, holidays, cultural celebrations or accounting periods. The cyclical component is similar to seasonal fluctuations, but the cycle period is not as regular as seasonality. That is, the periodicity is irregular. This makes the cyclical component difficult to predict. It is usually subjectively assessed. Thus, we will leave the forecasting of the cyclical component to the third course in this specialization. Generally, the economic cycle will influence the cyclical behavior of the series. The random component is non-systematic and cannot be predicted with any accuracy. Typically, the random component incorporates effects on the time series that cannot be explained by the variables that influence the systematic components. The random component can also include one off effects, such as the introduction of a new tax system or natural disasters. The extent of the random component will determine the maximum level of forecast accuracy that is achievable. Now that we know about the components, let's look at modeling. Although it is tempting to try and produce a model to explain or fit the observed time series sample, which we call within sample forecasts, forecasting requires the model to accurately predict another sample, the future which we call the out of sample forecasts. Thus, it is worthwhile to try to find a general model that will explain and predict both past and future samples. It is worthwhile to try and model the general process, which generates observations that is the population model as opposed to the sample model for the observed time series sample. Correctly specified models will explain the systematic component of the generating process. Forecasts from these models will, on average, be accurate, assuming zero mean error. The error term overall should be symmetric and around zero mean. In a correctly specified model, the variance of the error term should be smaller than for non-correct specifications. Typically, forecasters choose the model specification that provides unbiased forecasts and reduces error variance. Additionally, the errors are uncorrelated over time. Okay, what is this error term? The sample forecast error, also known as the residual, is equal to the actual data minus the forecaster data for each point in time. Plots of errors, i.e., residuals over time should reflect what I mentioned before. A mean of zero approximately equal numbers of residuals, either side of zero and no systematic patterns over time. This can be checked via an examination of the scatter plots of residuals versus time, or via a correlogram. The accuracy checks of our models are via error functions, which are mathematical summaries of the errors. The error functions are usually based on either the absolute errors or the squared errors. This is due to the misleading interpretation of the average of errors, which is going to be close to zero as the positive and negative errors cancel each other out. There are many error functions, and in this course, we focus on the mean absolute error, as well as the mean squared error. Now over to the Excel screen floor videos where we will work through the ideas introduced in this video. I encourage you to download the relevant Excel workbook and work alongside me as you watch each video. Please make sure that you attempt the quizzes to practice what you've learned and receive some feedback on your learning. You're soon going to be an expert at business forecasting. Everyone say wow let's jump to it. [MUSIC]
