Two types of data

Cross-sectional data are observations collected at a single point in time.

A time series consists of data that are collected, recorded, or observed over successive
increments of time.

1)Mean Stationary Data

If the data is generated by a constant process subject to random error, ie the mean and the
variance around the mean remains constant overtime then it is mean stationary. Non-
trending and non-seasonal.

2)Trend: The trend is the long-term component that represents the growth or decline in
the time series over an extended period of time.

3)Seasonality: The seasonal component is a pattern of change that repeats itself year after
year.

4)Cyclicity - The cyclical component is the wavelike fluctuations around the trend.
When observations exhibit rises and falls that are not of a fixed period, a cyclical
pattern exists. The cyclical component is the wavelike fluctuation around the trend that
is usually affected by general economic conditions. A cyclical component, if it exists,
typically completes one cycle over several years. Cyclical fluctuations are often influenced
by changes in economic expansions and contractions, commonly referred to as
the business cycles. Figure 2 also shows a time series with a cyclical component. The
cyclical peak at year 9 illustrates an economic expansion and the cyclical valley at year
12 an economic contraction.
When observations are influenced by seasonal factors, a seasonal pattern exists.
The

5)Irregularities: Part that cannot be forecasted.

Autocorrelation: Autocorrelation is the correlation between a variable lagged one or more
time periods and itself. For trend, for seasonal, for random data - autocorrelation values will
be different.

Correlogram: The correlogram or autocorrelation function is a graph of the autocorrelations

for various lags of a time series.

Naive model
Averaging methods:
Simple Average:

Single moving average:

Double moving average

Exponential Smoothing Methods:

1. Simple Exponential Smoothing Method:
2. Holt’s

ARIMA::
Decomposition:

Accuracy Measures:
MAPE - Mean absolute percentage Error

MAE/MAD - Mean Absolute Error/ Mean absolute Deviation

MSE - Mean Squared Error

RMSE - Root of Mean Squared Error

MPE - Mean Percentage Error

Session 6 7 8---------------------------------------------------------------------------------
data: use https://docs.google.com/spreadsheets/d/1mZgBvEO_fAbA-
pmW8JQtHB79pqJPODul2nerlXq74-M/edit?usp=sharing
Exponential Model Smoothening

Stage 3 - Estimates of trend and seasonal components, and parameter values are made at
this stage.

Stage 4: Iterative process. Because most optimal parameter values are unknown.
1. Naive model
a. Simple naive: Y^t+1 = Yt
Does not take into consideration the increase in trending data
b. Naive + Trend: Y^t+1= Yt+ (Yt-(Yt-1)) - it assumes that change is same as
that in last period
c. Naive + Seasonality = Y^t+1 = Yt-3
 d. Naive Trend + Seasonality Y^t+1 = Yt-3 + ((Yt- Yt-1) + (Yt-1 - Yt-2) + (Yt-2 -
Yt-3) + (Yt-3 - Yt-4))/4
= Yt-3 + (Yt - Yt-4)/4

2. Averaging Methods: Applying equal weights to all data points/observations taken

into considerations.

a. Simple Average (Mean): assume that all the data points are equally
important to forecast the next observation. Used for data that doesn’t have
trend or seasonality.

-----------------------------------------------------------------------------------------------------------------

The above equation can also be written as:

Since only the most recent forecasted data and recent observations need to be
stored, this method is used for inventory management - because a large number of
series are forecasted simultaneously.

b. Single Moving Average Method MA(k): For stationary data. If it is used for
trending data - projected value will be consistently low. (but will handle such
data better than simple average)

Recognizes that data points closer to the forecasting time period data points are more
important.
So it takes only a fixed number of data points to forecast a value - Order (k)
When using this method for directly forecasting - you always remove the oldest data points.
e.g . for the simple moving average of order 3, you will remove the oldest 2 data points.
1. Sudden significant change - smaller order moving average
2. Widespread change - higher-order moving average
 Eg. moving average of order 3. - you are losing out 2 data points - you leave out
the oldest 2 data points. k=3
K is used as the notation for the order.

the more observations included in the moving average, the greater the smoothing e®ect. A
forecaster must choose the number of periods (k) in a moving average. The two extreme
cases are k = 1 (naive model) and k = n (simple average)

a small value for k will allow the moving average to follow the pattern, but these MA
forecasts will nevertheless trail the pattern, lagging behind by one or more periods.

each new forecast (Ft+2) is simply

an adjustment of the immediately preceding forecast (Ft+1).

if k is a big number, this adjustment is small, so that moving

averages of high order provide forecasts that do not change very much, because the proportion of
new values taken to forecast is very low.

For eg. For MA(10), only 1 out of the 10 values will change in forecasting for 2 consecutive periods.

In conclusion, an MA(k) forecasting system will require k data

points to be stored at any one time.

What happens when MA or Simple Average is used to forecast trending and seasonal data
c.. Double moving average - for non-stationary trending data.

https://www.acq.osd.mil/dpap/cpf/docs/contract_pricing_finance_guide/vol2_ch6.pdf - from
Page 20

Double moving average lags behind Simple moving average as much as how much simple
moving average lags behind original.

When you add up the distance between the simple moving average and double moving
average to the simple moving average, it will give a value closer to actual value.

(Kuch Samaj Nahi Aya upper wala)

3. Exponential Smoothing Methods: unequal set of weights to past data. Applies weights
that decay exponentially from the most recent data to most distant data. Taking a weighted
average of past data points with weights that decay smoothly.

Need parameters to be defined - with values that lie between 1 and 0. - this will decide the
weights
Recursive in nature - moving through the known data period by period, as opposed to using
all the past data in one fitting exercise.
a. Single Exponential Smooth Method:
Only one parameter needs to be estimated.

------( 4.4)
where alpha is a constant between 0 and 1.

new forecast is simply the old forecast plus an adjustment for the error that occurred
in the last forecast.
If you take an alpha close to 1, a larger adjustment is made to the forecasted value
If you take an alpha close to 0, a smaller adjustment is made to the forecasted value

will always trail any trend in the actual data, since the most the method can do is
adjust
the next forecast for some percentage of the most recent error.

Solves any any storage problem, because it is no longer necessary to store all of the
historical data or a subset of them (as in the case of the moving average). Rather,
only the most recent observation, the most recent forecast, and a value for ® must
be stored.
One point of concern relates to the initializing phase of SES.
For example, to get the SES forecasting system started we need F1
because

Since the value for F1 is not known, we can use the first observed value (Y1) as the first
forecast (F1 = Y1)

the initial forecast F1 plays a role in all subsequent forecasts. But the
weight attached to F1 is usually small.

If a small value of alpha is chosen, the initial forecast plays a more prominent role than when
a larger alpha is used. Also, when more data are available t is larger and so the weight
attached to Ft is smaller.
 The smoothing effect of alpha can be seen in Figure 4-7. A large
value of alpha (0.9) gives very little smoothing in the forecast, whereas

a small value of alpha (0.1) gives considerable smoothing. When alpha = 1,

(4.4) shows that exponential smoothing is equivalent to using the last
observation as a forecast. That is, it is the same as NF1, the naÄve
forecast method 1.

b. Holt’s Linear Method:

Makes use of 2 different parameters.
Allows forecasting for trend series.

The forecast for Holt's linear exponential smoothing is found using two
smoothing constants, alpha and beta (with values between 0 and 1), and
three equations:
 If alpha and beta are closer to 1, very less smoothing will happen.

For first value, we take L1 = Y1, b1=0.

For choosing the best combination of alpha and beta, first decide alpha, then decide beta.

For ARIMA:
ARIMA(0,1,1)