Chapter 18 - Modelling Flashcards
What are the 2 main deterministic methods of testing the variability of model results?
Sensitivity analysis
Scenario testing
What is sensitivity analysis?
Varying the initial assumptions one by one and assessing the impact on the model output.
What is scenario testing?
Changing several assumptions in combination and assessing the impact on the model output.
What is a model?
A simplified version of reality which captures the essential features of a problem and aids our understanding.
What are some characteristics of a good model? (9)
Modelling process should be adequately documented.
Model should be understood and results must be communicable.
The model should exhibit sensible behaviour in light of model variables which are correlated.
The model results should be able to be verified independently for reasonableness.
The model should not be overly complex and difficult to interpret.
The model should be capable of development and refinement.
A range of methods of implementation should be available to facilitate testing, parameterisation and focus of results.
The frequency of model re-calibration will affect the reliability and speed of the model outputs.
What is a model point?
A representative single policy in a homogeneous group
What factors will determine the number of model points used?
Computing power Time constraints Heterogeneity of the class Sensitivity of the results to different model points The purpose of the model
What is a deterministic model?
Model where parameter values are fixed at the outset and a single outcome is obtained. Sensitivity analysis and scenario testing can be carried out to assess the variability of results.
What are the advantages and disadvantages of deterministic models?
Advantages:
Easier to communicate results and workings of the model
It explicitly tests different economic scenarios
The model is usually cheaper, easier to design and more efficient to run than a stochastic model
Disadvantages:
It requires expert judgement as to which scenarios will be tested, and our decisions are often biased and limited by our imagination.
May not effectively capture the variability of parameters in reality
What is a stochastic model?
Model where parameters are assigned a probability distribution. The model is run several times (simulation) and so a range of values are outputted, which gives an understanding of the potential distribution of results.
What are the advantages of stochastic models?
The model can test a wider range of economic scenarios.
Can produce quality results when data used to determine parameter distributions is appropriate.
Good for allowing for and quantifying uncertainty.
What are possible disadvantages of stochastic models?
More complex programming and therefore less efficient to run.
May introduce spurious accuracy
Increased difficulty in interpretation and communication of results
Questionable appropriateness of distribution of parameters.
What does it mean for a model to be dynamic?
Asset and liability parts of the model should interact similar to how they would in real life in different circumstances.
What are the steps involved when developing a deterministic model? (9)
Specify the purpose of the model
Collect, group and modify data
Choose the form of the model and identify parameters
Use past experience and estimation to assign parameter values
Construct a model that replicates expected cashflows
Test the model and try to identify errors in structure
Check an acceptable goodness of fit
Run the model using estimated future values of the parameters
Perform scenario and sensitivity testing and assess the robustness of the model
What are the steps in developing a stochastic model? (10)
Specify the purpose of the model
Collect, group and modify data
Choose the form of the model and identify parameters
Specify the correlation between variables
Use part experience and estimation to choose a suitable density function for each of the parameters to be modelled stochastically
Construct a model that replicates expected cashflows
Test the model and try to identify errors in structure
Check an acceptable goodness of fit
Run the model several times (simulation)
Produce a summary of the results that shows the distribution of the modelled results after many simulations have been run (confidence intervals)
