ARMA & ARIMA MODEL SHEET

ARMA – This model corrects predictions by learning from past forecasting

ARIMA – Predictions are based on past patterns.

Model Summary –
p – Number of autoregressive terms
q – Number of differencing steps
r – Number of moving average terms
ACF (Auto Correlation Plot)
1. Shows correlation between a time series & its lagged value.

2. For a good ARIMA value, residual should not have significant spikes
PACF (Partial Auto Correlation Function Plot)
1. Shows correlation between a time series and its lagged values after accounting for a correlation with
intermediate lag

2. Helps identify p (number of autoregressive terms) order

3. Significant spikes at initial lags suggest the order of AR terms.

Thing to look for in ARMA ARIMA Model

Favourable
Term Meaning Use
Value
Evaluate relative quality of Stats
AIC Akaike Information Criterion Lower
Model

BIC Bayesian Information Criterion Lower Same

R Square &
Higher How well the Model fits the Data
Adjusted R Square
Mean Absolute Percentage
MAPE Lower Measure the accuracy of Model
Error

MAE Mean Absolute Error Lower Absolute Terms

Log Likelihood Higher Indicates Model Fit

Sensitivity Report Analysis Sheet

 Favourable
Term Meaning Use
Value
The Maximum / Minimum Profit It indicates the best achievable
Objective Function Depends
achieved under the given constraints outcome based on the current model
These values show how resources should allocated to
The Values of the DV that yield
Decision Variable Depends achieve the objective functions
the optimal solution.
orignal value
The amount by which the objective
Reduced Cost
function coefficient of a non basic
(Shadow Price of
variable must improve before that
Decision Varibale)
variable enters the optimal solution
A +ve Positive SP indicates increasing the constraints
The change in objective functions value per
RHS will improve the objective Function
Shadow Price unit increase in the RHS of a constraint, Lower
assuming all else remains constant
A Zero shadow price means no impact
The range withing objective function can change Sensitivity of a model to changes in the
Allowable Increase/Decrease Lower
without altering the optimal solution profitability or cost of the decision variables.
The available resources or the limit set for a Represents the resource availability or a
Constraint RHS Depends
constraint boundary condition
Constraints that are exactly met in the These constraints are fully utilized and limit the
Binding Constraints
optimal solution(RHS = LHS) solution space (Slack = 0)
Constrains that are not fully uitilized in the These constraints have slack or surplus and do not
Non Binding Constraint
optimal solution restrict the soltuion
The unused portion of a resource in constraints
Slack Indicates the amount of resource that is still available
with a "less than or equal to" inequality
Surplus The Extra amount Represents the excess beyond the requiremtent
Relationship between the primal and derived Every constraint in the primal corresponds to a
Duality
problem. variable in the dual & vice versa
The allowable range for objective coefficient Shows the robustness of the optimal solution
Objective Coefficient Ranges
before the optimal basis changes to changes in costs or profits of the varibales
Decision Analysis –
Scene - A company is deciding whether to launch Product A, Product B, or do nothing, depending on
market conditions (High Demand or Low Demand).
Decision High Demand Low Demand
Product A 100 20
Product B 80 30
Do Nothing 0 0

High Demand: It represents the profit for each decision if market condition are favourable.
Low Demand: It represents the profit for each decision if market condition are unfavourable.
Regret = Best Payoff – Chosen Payoff

Regret table
Decision High Demand Low Demand
Product A 0 10
Product B 20 0
Do Nothing 100 30

In this the best pay off for high demand is 100, so we subtracted everything from 100. For low demand the
best pay off is 30.

Adding Probabilities
We will assume the following probabilities for the market demand

High Demand = 60%

Low Demand = 40%

With these probabilities we can calculate Expected Monetary Value (EMV)

EMV of a product = Sum of (Payoff x Probability)

Decision High Demand (Payoff x Probb) Low Demand (Payoff x Probb) EMV
Product A 100 x 60% = 60 20 x 40% = 8 60 + 8 = 68
Product B 80 x 60% = 48 30 x 40% = 12 48 + 12 = 62
Do Nothing 0 0 0

With these probabilities we can calculate Expected Opportunity Loss (EOL)

EOL of a product = Sum of (Regret x Probability)
 Decision High Demand (Payoff x Probb) Low Demand (Payoff x Probb) EOL
Product A 0 x 60% = 0 10 x 40% = 4 4
Product B 20 x 60% = 12 0 x 40% = 0 12
Do Nothing 100 x 60% = 60 30 x 40% = 12 72

FINAL TABLE WITH ALL THE VALUES

Decision EMV EOL
Product A 68 4
Product B 62 12
Do Nothing 0 72

Making Recommendations –
1. EMV Criterion – Choose the decision with the highest EMV that is product A (68)
2. EOL Criterion – Choose the decision with the lowest EOL that is product A (4)
The product A has the best decision considering both EMV and EOL approaches.

Maximax & Maximin approach –

Criterion Uses Probabilities? Focus Risk Type
EMV Yes Weighted average of payoffs Risk-neutral
EOL Yes Weighted average of regrets Risk-neutral
Maximax No Maximum possible payoff Risk-seeking
Maximin No Maximum worst-case payoff Risk-averse

Maximax – selects the decision with the highest possible payoff. (Optimistic)
Maximin – selects the decision with the highest worst-case payoff. (Pessimistic)

Decision Tree - It is a graphical representation of decision, possible outcomes and their probabilities it helps
to analyse decision making problems systematically particularly in complex situation with uncertainties
1. Decision Node (Square) – They represent a decision point where you choose an action.
 2. Chance Node (Circle) – Represents an event with uncertain outcomes.

3. Branches – Extend from nodes to show actions or outcomes

Let us start with basic Decision Tree –