What can be forecasted easily?
daily electricity demand in 3 days’ time
Google stock price tomorrow
Google stock price in 6 months’ time
maximum temperature tomorrow
total sales of drugs in Australian pharmacies next month
Something is easy to forecast if:
we have a good understanding of the factors that contribute to it
there is a lot of data available
the future is somewhat similar to the past
ID assumption: samples are identically distributed
the forecasts cannot affect the thing we are trying to forecast.
self-fulfilling prophecies (election polls)
controlled systems
Dealing with uncertainty
Aleatoric (alea=latin for dice) or data uncertainty. This is the irreducible uncertainty that we cannot reduce by increasing observations.
simple case: homoscedastic (noise with fixed distribution)
more realistic case: heteroscedastic
Epistemic (epistḗmē=science) or model uncertainty. This uncertainty can be reduced by increasing observed data, and as such, can be reduced
Aleatoric or epistemic uncertainty?
Another source of uncertainty: model choice
Model choice is another source of uncertainty, that can be mitigated by model selection. Unfortunately, it cannot correct ideological biases.
The M6 financial forecasting competition: 03-2022/ 01-2023
Its purpose is to shed new light on the EMH (Efficient Market Hypothesis) by explaining the poor performance of professionally managed funds
The competition committee made the following hypothesis:
There will be a weak link between the ability of teams to accurately forecast individual rankings of assets and risk-adjusted returns on investment
Decision-making and unforecastable signals
If some conditions are met, we can make informative decisions even if the underlying processes are not forecastable. Imagine you are a trader, which of the following signals you would like to forecast?
A pseudo-random binary signal
A noisy signal
One lucky realization
