VoseTimeSME2Perc

VoseTimeSME2Perc({Percentiles1}, {Percentiles2}, P1, P2, Correlation Factor, Negative Allowed)

Time series function modeling a variable estimated for each period by a lower and upper percentile.

• {Percentiles1} is an array of values of the P1 percentile in each period of the forecast.

• {Percentiles2} is an array of values of the P2 percentile in each period of the forecast.

• P1 is the probability used together with {Percentiles1}. For example, if P1 is set to 10% the {Percentiles1} values are interpreted as the values for which, in each individual period, the variable has a 10% probability of being below. P1 must lie on [0,1].

• P2 is the probability used together with {Percentiles2}. For example, if P2 is set to 90% the {Percentiles2} values are interpreted as the values for which, in each individual period, the variable has a 90% probability of being below. P2 must lie on [0,1].

• Correlation Factor applies a positive correlation between generated values within each period of the series. CorrelationFactor must lie on [0,1]. Optional, set to zero if omitted.

• Negative Allowed is a Boolean parameter specifying whether the series may take negative values (Negative Allowed = TRUE) or not (Negative Allowed = FALSE). This allows the user to avoid a common problem when estimating with percentiles that the resultant distribution can extend beyond plausible values.

Explanation and Uses

The SME2Perc time series function provides an easy, subjective way to specify a time series with some key features:

Growth and spread over time can be controlled by changing the {Percentiles1} and {Percentiles2} array values. P1 and P2 would most commonly be set at {0.2, 0.8}, {0.1, 0.9} or {0.05, 0.95} reflecting 1 in 5, 1 in 10 and 1 in 20 probabilities respectively, which are probabilities that people can realistically understand. Avoid values like {0.01, 0.99} or more extreme if possible, because human beings are not that great at appreciating and estimating probabilities with that level of precision.

Correlation between periods can be specified with a single parameter. The level of correlation is best selected by reviewing the example pathways that are generated in the interface each time one clicks the Generate button. Look at the range of variation from one period to the next across the entire series and adjust the CorrelationFactor until it looks reasonable. If you believe that there is correlation across the series you will likely settle on a value above 0.4 since lower levels of correlation are not immediately obvious to the eye. You will want to use correlation, for example, when the variable being forecast will tend to be high in each year if it is high in the first year: for example, a forecast of sales of a new product, when it either takes off because it is appealing to potential clients, or doesn’t; or the use of a new vaccine where people are generally convinced of its value, or not.