TIME SERIES FORECASTING

Time Series
A time series is a sequence of observations on a variable measured at successive points
in time or over successive periods of time. The measurements may be taken every hour,
day, week, month, or year, or at any other regular interval.
The pattern of the data is an important factor in understanding how the time series has
behaved in the past. If such behavior can be expected to continue in the future, we can
use the past pattern to guide us in selecting an appropriate forecasting method.
To identify the underlying pattern in the data, a useful first step is to construct a time
series plot. A time series plot is a graphical presentation of the relationship between
time and the time series variable; time is on the horizontal axis and the time series values
are shown on the vertical axis. Let us review some of the common types of data patterns
that can be identified when examining a time series plot.
Components of a Time Series
A time series, in general, is supposed to be affected by four main components also
known as the time series patterns, which can be separated from the observed data. These
components are Trend, Cyclical, Seasonal and Irregular components. A brief
description of these four components is given here.
Trend pattern
The general tendency of a time series to increase, decrease or stagnate over a long period
of time is termed as Secular Trend or simply Trend. Thus, it can be said that trend is a
long-term movement in a time series (Year(s)). A trend is usually the result of long-
term factors such as population increases or decreases, changing demographic
characteristics of the population, technology, and/or consumer preferences.
Seasonal pattern
Seasonal variations in a time series are fluctuations within a year during the season. The
important factors causing seasonal variations are: climate and weather conditions,
customs, traditional habits, and so on. For example, sales of ice-cream increase in the
dry season, sales of woolen cloths increase in the rainy season. Seasonal variation is an
important factor for businessmen, shopkeeper and producers for making proper future
plans.
Cyclical pattern
The cyclical variation in a time series describes the medium-term changes in the series,
caused by circumstances, which repeat in cycles. The duration of a cycle extends over
a longer period of time, usually two or more years. Most of the economic and financial
time series show some kind of cyclical variation. For example, a business cycle consists

1|Page
of four phases, which include; Prosperity, Decline, Depression and Recovery.
Schematically a typical business cycle can be shown as below:

Figure 1: A Four-Phase Business Cycle

Irregular pattern
Irregular or random variations in a time series are caused by unpredictable influences,
which are not regular and also do not repeat in a particular pattern. These variations are
caused by incidences such as war, strike, earthquake, flood, revolution, and so on. There
is no defined statistical technique for measuring random fluctuations in a time series.
Considering the effects of these four components, two different types of models are
generally used for a time series viz. Multiplicative model and Additive model.
Multiplicative Model : Y (t) = T (t)× S(t)×C(t)× I(t)
Additive Model : Y (t) = T (t) + S(t) + C(t) + I(t)
Here Y(t) is the observation and T(t), S(t), C(t) and I(t) are respectively the trend,
seasonal, cyclical and irregular variation at time t.
Multiplicative model is based on the assumption that the four components of a time
series are not necessarily independent and they can affect one another; whereas in the
additive model, it is assumed that the four components are independent of each other.
Horizontal Pattern
A horizontal pattern exists when the data fluctuate around a constant mean. Changes in
business conditions can often result in a time series that has a horizontal pattern shifting
to a new level. For instance, suppose the gasoline distributor (Tradex Etoug-ebe signs
a contract with the Etoug-Ebe Council to provide gasoline for state cars located in
Etoug-ebe. With this new contract, the distributor expects to see a major increase in
weekly sales starting from the next week.

2|Page
Forecasting
A forecast is a prediction of some future event or events. As suggested by Neils Bohr,
making good predictions is not always easy. The main aim of time series modeling is
to carefully collect and rigorously study the past observations of a time series to develop
an appropriate model that describes the inherent structure of the series. This model is
then used to generate future values for the series; that is, to make forecasts.
Time series forecasting thus can be termed as the act of predicting the future by
understanding the past. Due to the indispensable importance of time series forecasting
in numerous practical fields such as business, economics, finance, science, and
engineering, amongst others; proper care should be taken to fit an adequate model to
the underlying time series
Forecast Uses
 Forecasts are useful for planning activities in the operations, marketing,
financial, and perhaps all other functions of a business.
 The time horizon of a forecast indicates how far into the future it predicts.
 The required time horizon is one of the determinants of which approach to use
in forecasting.
The Forecasting Process
The following can be considered the five steps involved in the forecasting process.
Step 1: Collect numerical data from internal and external sources. Internal sources
may include previous sales figures, backorder logs, and production/inventory quantities.
External sources may include market-wide sales figures, national economic indicators,
and changes in customer demographics.
Step 2: Generate a forecast based on numerical data. Based on the time series data
and its determined pattern, we perform the forecast using the particular forecasting
technique which suites the available data.
Step 3: Check the accuracy of the forecasting technique. Not only should the
forecasting technique be chosen wisely, but the forecasting results should be monitored
on an ongoing basis to assure that the forecasts maintain their integrity.
Step 4: Include qualitative judgments not represented in the numerical data.
Sometimes the decision-maker has reason to believe that the future will be somewhat
different from the forecast because he or she has beliefs that are not represented in the
numerical data. For this reason, the forecast values may be adjusted to take into
consideration these subjective judgments.
Step 5: Apply the forecast in making decisions. A useful forecast not only includes an
expected outcome, but also a measure of the uncertainty about the point forecast. Basing
the decision only on the forecasted value (and not the uncertainty), can cause serious

3|Page
problems, particularly if the cost of the actual outcome exceeding the forecast is quite
different from the cost of it being lower than the forecast.
Time Series Forecasting Methods
Moving Averages
The moving averages method uses the average of the most recent k data values in the
time series as the forecast for the next period. Mathematically, a moving average
forecast of order k is as follows:
Moving Average Forecast of Order k
∑(𝑚𝑜𝑠𝑡 𝑟𝑒𝑐𝑒𝑛𝑡 𝑘 𝑑𝑎𝑡𝑎 𝑣𝑎𝑙𝑢𝑒𝑠) 𝑌𝑡 + 𝑌𝑡−1 + ⋯ + 𝑌𝑡−𝑘+1
𝐹𝑡+1 = =
𝑘 𝑘
Where
𝐹𝑡+1 = 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑖𝑒𝑠 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡 + 1
𝑌𝑡 = 𝑎𝑐𝑡𝑢𝑎𝑙 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑖𝑒𝑠 𝑖𝑛 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
To use moving averages to forecast a time series, we must first select the order, or
number of time series values, to be included in the moving average. If only the most
recent values of the time series are considered relevant, a small value of k is preferred.
If more past values are considered relevant, then a larger value of k is better.
Weighted Moving Averages
The weighted moving averages involve selecting a different weight for each data value
in the time series and then computing a weighted average of the most recent k values as
the forecast unlike in the moving averages method, each observation in the moving
average calculation receives the same weight. In most cases, the most recent observation
receives the most weight, and the weight decreases for older data values. It is equally
worth noting that for the weighted moving average method the sum of the weights is
equal to One (1).
To use the weighted moving averages method, we must first select the number of data
values to be included in the weighted moving average and then choose weights for each
of the data values. In general, if we believe that the recent past is a better predictor of
the future than the distant past, larger weights should be given to the more recent
observations. However, when the time series is highly variable, selecting approximately
equal weights for the data values may be best. The only requirement in selecting the
weights is that their sum must equal 1.
To determine whether one particular combination of number of data values and weights
provides a more accurate forecast than another combination, we recommend using MSE
as the measure of forecast accuracy. That is, if we assume that the combination that is
best for the past will also be best for the future, we would use the combination of number

4|Page
of data values and weights that minimizes MSE for the historical time series to forecast
the next value in the time series.
Exponential Smoothing
This method also uses a weighted average of past time series values as a forecast; it is
a special case of the weighted moving averaged method in which we select only one
weight. The weight for the most recent observation. The weights for the other data
values are computed automatically and become smaller as the observations move
farther into the past. The exponential smoothing equation is as follows;
Exponential Smoothing Forecast
𝐹𝑡+1 = 𝛼𝑌𝑡 + (1 − 𝛼)𝐹𝑡
Where
𝐹𝑡+1 = 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑖𝑒𝑠 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡 + 1
𝑌𝑡 = 𝑎𝑐𝑡𝑢𝑎𝑙 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑖𝑒𝑠 𝑖𝑛 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝐹𝑡 = 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑖𝑒𝑠 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝛼 = 𝑠𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (0 ≤ 𝛼 ≤ 1)
This method is similar to the weighted moving average method in that you can
experiment with different values of α and choose the value that provides the smallest
forecast error using one or more of the measures of forecast accuracy (MAE, MSE, or
MAPE).
Linear Trend Regression
Using this method, we declare the estimated regression equation describing a straight-
line relationship between an independent variable x and a dependent variable y which
is written as;
𝑦̂ = 𝛽0 + 𝛽1 𝑥
Where 𝑦̂ is the predicted value of y. To emphasize the fact that in forecasting the
independent variable is time, we will replace x with t and with Tt to emphasize that we
are estimating the trend for a time series. Thus, for estimating the linear trend in a time
series we will use the following estimated regression equation.
Linear Trend Equation
𝑇𝑡 = 𝛽0 + 𝛽1 𝑡
Where
𝑇𝑡 = 𝑙𝑖𝑛𝑒𝑎𝑟 𝑡𝑟𝑒𝑛𝑑 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑖𝑛 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝛽0 = 𝑖𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑙𝑖𝑛𝑒𝑎𝑟 𝑡𝑟𝑒𝑛𝑑 𝑙𝑖𝑛𝑒
𝛽1 = 𝑠𝑙𝑜𝑝𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑙𝑖𝑛𝑒𝑎𝑟 𝑡𝑟𝑒𝑛𝑑 𝑙𝑖𝑛𝑒
𝑡 = 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑

5|Page
Formulas for computing the estimated regression coefficients (b1 and b0) are as
follows;
Computing the Slope and Intercept for a Linear Trend
∑𝑛𝑡=1(𝑡 − 𝑡̅)(𝑌𝑡 − 𝑌̅)
𝛽1 =
∑𝑛𝑡=1(𝑡 − 𝑡̅)2
𝛽0 = 𝑌̅ − 𝛽1 𝑡̅
Where
𝑌1 = 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑖𝑒𝑠 𝑖𝑛 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡
𝑛 = 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑖𝑚𝑒 𝑝𝑒𝑟𝑖𝑜𝑑𝑠 (𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠)
𝑌̅ = 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑒𝑟𝑜𝑒𝑠
𝑡̅ = 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑡
Nonlinear Trend Regression
In some cases, the time series plot indicates an overall increasing or upward trend. But
the linear trend may not appear to be appropriate. Instead, a nonlinear or curvilinear
function appears to be needed to model the long-term trend. A variety of nonlinear
functions can be used to develop an estimate of the trend for the time series values. For
instance, consider the following quadratic trend equation:
𝑇𝑡 = 𝛽0 + 𝛽1 𝑡 + 𝛽2 𝑡 2
Where t = 1 corresponds to year 1, t = 2 corresponds to year 2, and so on. Due to the
complexity of manually computing forecasts using this method various businesses turn
to use software such as MS Excel to perform such operations and are especially used to
develop estimates of β0, β1, and β2 for a quadratic trend equation.
Seasonality and Trend
Here, we look at how to develop forecasts for time series that has a seasonal pattern. By
examining the extent to which seasonality exists, we incorporate it into our forecasting
models to ensure accurate forecasts. We begin by considering a seasonal time series
with no trend.

Seasonality Without Trend

Using the sample time series data on the table below;

6|Page
 Table 1: Umbrella Sales Time Series
Year Quarter Sales
1 1 125
2 153
3 106
4 88
2 1 118
2 161
3 133
4 102
3 1 138
2 144
3 113
4 80
4 1 109
2 137
3 125
4 109
5 1 130
2 165
3 128
4 96
Considering the example on the table above, the time series plot (when plotted) does
not indicate any long-term trend in sales. Unless the data is looked at carefully, you
might conclude that the data follow a horizontal pattern and that single exponential
smoothing could be used to forecast sales. But closer inspection of the time series plot
reveals a pattern in the data. That is, the first and third quarters have moderate sales, the
second quarter has the highest sales, and the fourth quarter tends to be the lowest quarter
in terms of sales volume. Thus, we would conclude that a quarterly seasonal pattern is
present.
We can treat the season as a categorical variable and as such make use of dummy
variables which is the most appropriate approach and generally when a categorical
variable has k levels, k-1 dummy variables are required. For instance, in the umbrella
sales time series season is a categorical variable with four levels: quarter 1, quarter 2,
quarter 3, and quarter 4. Thus, to model the seasonal effects in the umbrella time series
we need 4 - 1 = 3 dummy variables. The three dummy variables can be coded as follows:
1 𝑖𝑓 𝑄𝑢𝑎𝑟𝑡𝑒𝑟 1 1 𝑖𝑓 𝑄𝑢𝑎𝑟𝑡𝑒𝑟 2 1 𝑖𝑓 𝑄𝑢𝑎𝑟𝑡𝑒𝑟 3
𝑄𝑡𝑟1 = { 𝑄𝑡𝑟2 = { 𝑄𝑡𝑟3 = {
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Using 𝑌̂ to denote the estimated or forecasted value of sales, the general form of the
predicted regression equation relating the number of umbrellas sold to the quarter the
sales take place follows:
𝑌̂ = 𝛽0 + 𝛽1 𝑄𝑡𝑟1 + 𝛽2 𝑄𝑡𝑟2 + 𝛽3 𝑄𝑡𝑟3

7|Page
Using the multiple regression equation to perform the forecasts manually is very
complex thus resulting in the use of software to calculating the coefficients of intercept
and slopes of the equation before calculation of the various forecasts for the respective
quarters. However, it is worth noting that we could/can obtain the quarterly forecasts
for the next year simply by computing the average number of umbrellas sold in each
quarter, as shown in the following table.
Forecast for Year 6
Year Qtr1 Qtr2 Qtr3 Qtr4
1 125 153 106 88
2 118 161 133 102
3 138 144 113 80
4 109 137 125 109
5 130 165 128 96
Average 124 152 121 95

Nonetheless, using the regression equation output provides additional information that
can be used to assess the accuracy of the forecast and determine the significance of the
results. And, for more complex types of problem situations, such as dealing with a time
series that has both trend and seasonal effects, this simple averaging approach will not
work.
Forecasts Accuracy
This involves several measures of forecast accuracy. These measures are used to
determine how well a particular forecasting method is able to reproduce the time-series
data that are already available. By selecting the method that has the best accuracy for
the data already known, we hope to increase the likelihood that we will obtain better
forecasts for future time periods.
The key concept associated with measuring forecast accuracy is “forecast error”, and is
defined as:
Forecast Error = Actual Value - Forecast
Using this formula, if the forecast error result is positive, it indicates that the business
underestimated the actual value of the time series (sales, stocks and so on). And if the
forecast error result is negative, it indicates that the business overestimated the actual
value of the time series. Thus, the forecast error may be positive or negative, depending
on whether the forecast is too low or too high.

The various techniques to measuring forecast accuracy include the following;

8|Page
Mean Absolute Error (MAE)
The mean absolute error, denoted MAE, is the average of the absolute values of the
forecast errors and is a measure of forecast accuracy that avoids the problem of positive
and negative forecast errors offsetting one another. MAE can be computed as follows;
𝐴𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟
𝑀𝐴𝐸 =
𝑛−1
Mean Squared Error (MSE)
This is another measure that avoids the problem of positive and negative forecast errors
offsetting each other. It is the average of the sum of squared forecast errors. This
measure of forecast accuracy computed as follows;
∑ 𝑠𝑞𝑢𝑎𝑟𝑒𝑑 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟
𝑀𝑆𝐸 =
𝑛−1
Mean Absolute Percentage Error (MAPE)
The size of MAE and MSE depends upon the scale of the data. As a result, it is difficult
to make comparisons for different time intervals, such as comparing a method of
forecasting monthly sales to a method of forecasting weekly sales or to make
comparisons across different time series. To make such comparisons we need to work
with relative or percentage error measures. The mean absolute percentage error denoted
MAPE, is such a measure used to test forecast accuracy in such a case.
𝐴𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟𝑠
𝑀𝐴𝑃𝐸 =
𝑛−1
Conclusion
Will like to emphasize that there is no forecasting method which can be used in all cases
and also none which can provide the exact numbers for the next periods reason for
which they called forecast and not exact values. Thus, we have to select our technique
following the forecasting process minimize the risk of over- and/or under-estimating
though will always exist.

Application 1:

9|Page
Based on the table below and using the naive (most recent observation) forecasting
method, we obtained the following measures of forecast accuracy:
Table 2: Umbrella sales for ShopAll ltd
Absolute value Squared Absolute Value
Sales Forecast Percentage
Week Forecast of Forecast Forecast of Percentage
(000) Error Error
Error Error Error
1 17
2 21 17 4 4 16 19.05 19.05
3 19 21 -2 2 4 -10.53 10.53
4 23 19 4 4 16 17.39 17.39
5 18 23 -5 5 25 -27.78 27.78
6 16 18 -2 2 4 -12.50 12.50
7 20 16 4 4 16 20.00 20.00
8 18 20 -2 2 4 -11.11 11.11
9 22 18 4 4 16 18.18 18.18
10 20 22 -2 2 4 -10.00 10.00
11 15 20 -5 5 25 -33.33 33.33
12 22 15 7 7 49 31.82 31.82
Total 5 41 179 1.19 211.69

𝐴𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟 41

𝑀𝐴𝐸 = = = 3.73
𝑛−1 11
∑ 𝑠𝑞𝑢𝑎𝑟𝑒𝑑 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟 179
𝑀𝑆𝐸 = = = 16.27
𝑛−1 11
𝐴𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟𝑠 211.69
𝑀𝐴𝑃𝐸 = = = 19.24%
𝑛−1 11
Application 2:
Exercise 1: Bicylce Sales for ABC Sarl Exercise 2: Tradex Gasoline Sales
Week Sales(1000s)
Year Sales (000,000s)Yt
1 170
1 21.6 2 210
2 22.9 3 190
3 25.5 4 230
4 21.9 5 180
5 23.9 6 160
7 200
6 27.5 8 180
7 31.5 9 220
8 29.7 10 200
9 28.6 11 150
10 31.4 12 220

References

10 | P a g e
5 Statistical Methods For Forecasting Quantitative Time Series | Bista INC. (2016, May
31). Retrieved December 6, 2019, from Bista Solutions: Best Odoo
Implementation Company | Odoo Gold Partners website:
https://www.bistasolutions.com/resources/blogs/5-statistical-methods-for-
forecasting-quantitative-time-series/
6.4.1. Definitions, Applications and Techniques. (n.d.). Retrieved December 6, 2019,
from https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc41.htm
Burba, D. (2019, October 3). An overview of time series forecasting models. Retrieved
December 6, 2019, from Medium website: https://towardsdatascience.com/an-
overview-of-time-series-forecasting-models-a2fa7a358fcb
Daitan. (2019, October 11). Fundamentals of Time Series Data and Forecasting.
Retrieved December 6, 2019, from Medium website: https://medium.com/better-
programming/fundamentals-of-time-series-data-and-forecasting-15e9490b2618
How BI Tools Can Help To Improve Sales And Marketing Strategy? (2017, September
19). Retrieved December 6, 2019, from Bista Solutions: Best Odoo
Implementation Company | Odoo Gold Partners website:
https://www.bistasolutions.com/resources/blogs/how-bi-tools-can-help-to-
improve-sales-and-marketing-strategy/
Nair, A. (2019, August 19). An Introductory Guide To Time Series Forecasting.
Retrieved December 6, 2019, from Analytics India Magazine website:
https://analyticsindiamag.com/an-introductory-guide-to-time-series-
forecasting/
The Advantages of the Time Series Method of Forecasting. (n.d.). Retrieved December
6, 2019, from Bizfluent website: https://bizfluent.com/info-8610535-
advantages-time-series-method-forecasting.html

11 | P a g e