Methods in PRA I: Trends and Forecasting

Question 1

What are the 3 key points in how the past is related to the future in PRA trends and forecasting methods?

Answer 1

• The past is used as a model for the (only)
future - key that it is concerning only one future.
• The substantive jump from probability to
prediction requires caution - probability points to infinite amount of other possibilities, the selection therefore needs to be done carefully - quantitative issues.
• Triangulation and aggregation considerably
improves prediction accuracy - combination of different methods.

Question 2

Is it better to stick to one method or use several methods in forecasting?

Answer 2

Better results come from combining methods.

Question 3

What is the difference between prediction and explanation?

Answer 3

Prediction is pre hoc, explanation is post hoc. E.g. you can explain complex systems that cannot be predicted.
• Many valid explanations have no predictive power. Some predictions have no explanatory power (e.g. correlations, time series, simple arithmetic).

Question 4

What is the problem with prediction?

Answer 4

Predictions are not demonstrative in an ontological sense.
• Even when predictions are theoretically and empirically strong, they fail.
• Conversely, predictive success is not a ‘confirmation’ of the explanatory power as other theories may replace them later.

Question 5

What are the conditions necessary for prediction?

Answer 5

The past is a model for the future. All methods of rigorous prediction are based on pattern fitting and projection:
1.Data availability
2.Pattern discernability
3.Pattern stability
It requires defining a ‘normal’ condition to judge whether a particular pattern is or deviates from the ‘normal’ state of things.
(Critique: think of KPE Quantification and Politics - selection of what to measure, and in this sense what is ‘normal’, is political choice).

Question 6

How is prediction argumentative?

Answer 6

The ‘substantive jump’ (Rescher 1998): it is a move from premise that A can happen to A will happen - this does not occur without decision-making (it is an argumentative way of thinking).

Question 7

What is the relationship between prediction and uncertainty?

Answer 7

Prediction requires some uncertainty to be a true prediction (Rescher 1998). Otherwise it is merely a calculation, if we know all the facts.

Question 8

What is ‘regression to the mean’?

Answer 8

When something distributes itself around the mean - general truth to system, is that developments will tend to revert to the mean e.g. financial stocks can sky-rocket or plunge, but they will stabilise to the ‘normal’ over time.

Question 9

What does ‘self-fulfilling prophecy’ refer to in prediction?

Answer 9

It is a feed-back phenomena - often in economic system e.g. stock market, but can also be seen in political systems. Prediction causes reaction - we therefore have to adjust our prediction in a reflexive way BUT this requires us to predict how people will react to our predictions = never ending cycle. See Rescher 1998:50-51.

Question 10

How is prediction related to insurance logic (economic approaches)?

Answer 10

The most frequently used prediction methods are based on the assumption that we can learn from the past/ history.

Question 11

What are the 2 common categories for predictive methods?

Answer 11

1. Judgemental methods.

2. Formal methods.

Question 12

What are judgemental methods in prediction?

Answer 12

Wisdom of the crowds: linkage connecting claim to data: judgemental or intuitive process of competent/authoritative experts, aggregation of expert opinions.

Question 13

What is the problem with judgemental methods?

Answer 13

Social interaction can form bias and regression to the mean.

Question 14

What are 4 examples of judgemental methods?

Answer 14

1. Delphi method (Rand corporation): collecting results from individual experts without them knowing about each other - the law of averages may be true at times but some individual predictions will at times outshine the average (sometimes the radical claims are the true ones).
2. Prediction markets: contains the idea of advantage - the prediction is constantly revised as new information emerges; law of supply & demand; can be used to see if the market is well calibrated for predictions over time; based on the assumption that more people have better information than individual experts - conditions:
   • relevant, available and useful information to aggregate.
   • diversity of opinion and strong motivation to trade.
   • clear and easily adjudicated contracts (unambiguous outcomes).
3. Volatility index: useful for very short term predictions.
4. The good judgement project: program collecting individuals talented in prediction, making them forecasting experts - supported by the intelligence community in USA
   Program for the forecasters:
   • Gets feedback on performance over time
   • Some training in probability
   • Work in teams and prove yourself wrong
   • Removed from decision-making or other biases

Question 15

What are formal methods in prediction?

Answer 15

Linkage claim to data: inferential principles, theories or laws (mathematical laws and theories).

Question 16

What is the problem with formal methods?

Answer 16

Statistical data is not enough to make a prediction, it also requires e.g. context based knowledge. Problem of regression to the mean. Most central problem is defining what is normal.

Question 17

What are 3 examples of formal methods?

Answer 17

1. Trend extrapolation: premise that trends (time-series)
   continue in a linear fashion.
2. Curve fitting: non-linear trends (exponential, logit, sinusoidal etc.). Aims to find the simplest form that fits data.

1&2 e.g. Econometrics: conditional forecast, or try to estimate the independent variable. BUT the regression is only a probability, not the prediction itself (this is often misunderstood).

3. Models: combines analogy with structural/behavioural laws to form a representation of the issue. - e.g. Cox Proportional Hazard Model:
   • Used to estimate the likelihood of an event occurring at time ‘t’ (formula), provided that it has not occurred before.
   • Popular in estimating the likelihood of civil war conditioned on a range of background factors.
   -BUT does not take into account the relationship over time.

Question 18

What is retrodiction?

Answer 18

Looking from the present to the past - can lead us to draw assumptions about how/why an event occurred. This historical tracing can then be added to background knowledge for future predictions. BUT problem that this can lead to bias, which can lead to deterministic oversimplification of past event-chains, leading to false further future assessments. See Aligica, 2003.

Question 19

What is the importance of prediction in risk analysis?

Answer 19

Prediction allows risk analysis to build its theories on a rational, evidential basis (quantitative - either judgemental or formal - or both) and narrows down the possibilities of the future to create a specific outcome (one specified future) subsequently facilitating policy-making.

Question 20

What is the difference between a forecast and a prediction?

Answer 20

A forecast is basically a very specific prediction that could happen or not happen at a specific time. See Rescher, 1998.