4 ms·

Some implications of the Bank of England comparing itself to the MetOffice

26 points by mr_tyzic 10y ago

Normal_gaussian 10y ago
The largest problem with forecasting in economics is surely that others are also forecasting and attempting to manipulate the system. With weather forecasting the result doesn't change just because it can be calculated.
- matt4077 10y ago
  It's not so much "trying to manipulate" (in the malicious sense) but that economic forecasts tend to be self-fulfilling prophecies. If you think a stock is undervalued you'll buy it , which will raise the price. If you think the economy will grow, you'll invest, which will cause the economy to grow. The result is the "efficient market hypothesis" which posits that all public information is already reflected in market prices to the best of our abilities (with a bit of swarm intelligence and darwinian selection making it unlikely that an individual can beat the market on ability). All that's usually applied to prices but other indicators like unemployment may simply correlate to such a degree that it extends to them as well.
  - smallnamespace 10y ago
    It's usually applied to prices because the assumption is that the act of trading on perceived incorrect prices causes the price to converge to the market's expectations. Whereas for employment, you would need a mechanism that allows for 1) a public prediction of what the future unemployment path is and 2) a way for people to put money down if they disagree, such that it actually directly impacts unemployment.
  - Normal_gaussian 10y ago
    I didn't mean manipulate in a malicious sense, but that doesn't really matter. You have a good point in that economic forecasts tend to be self-fulfilling, and that it is unlikely an individual can beat the market just with economic plays. Of course players don't just have traditional economic moves at their disposal, political power translates pretty well into economic manipulation - and that isn't just effective at a large scale.
nl 10y ago
I don't get it. Of course the economy is non-deterministic. That doesn't mean that useful forecasts can't be made. Statistical models are used because of this. I agree of course that deriving and presenting a single forecast from a statistical model is a dangerous process. Perhaps this was the link the author was portraying by their quest for an Isobar map rather than just "Rain/Non rain". (Also, I see others are looking at "economic forecasting" as price forecasting. I read this more as macro-economic forecasting, in which the outcomes are much to manipulate)
- neximo64 10y ago
  The problem is you can make forecasts but they're just guesses at best. The basic model for forecasts for example underestimates the occurrence of black swan events, plain and simple. That's why these days they use the stress test approach, assume the worst market conditions then see if a bank can cope.
  - nl 10y ago
    The problem is you can make forecasts but they're just guesses at best. This isn't true. The current US unemployment rate is 4.7%[1]. It's pretty easy to look at the time series data, and come up with a likely range for the next reading. The possibility of black swan events (as well as inadequacies of models, measurement inaccuracies etc) mean this number isn't a sure thing: instead a statistical model provides a distribution of probabilities. For example, the probability distribution curve might give an 90% chance of the unemployment rate being between 4.5% and 4.9%. If we are trying to predict a year ahead, there much less certainty. Black swan events are just one of the causes of that, but it's still possible to forecast that the unemployment rate in December 2017 has a (say) 30% chance of being between 5.5% and 4.5%. This isn't guesses (well to be clear: I'm guessing above). You model the inputs, and build a Bayesian belief network (say) to model the interactions between events. Here's a version of The Fed's model: https://www.federalreserve.gov/econresdata/notes/feds-notes/2014/a-tool-for-macroeconomic-policy-analysis.html https://www.federalreserve.gov/econresdata/notes/feds-notes/.... That's why these days they use the stress test approach, assume the worst market conditions then see if a bank can cope. That's completely different to macro-economic forecasting. [1] https://data.bls.gov/timeseries/LNS14000000 https://data.bls.gov/timeseries/LNS14000000

26 points by mr_tyzic 10y ago

The largest problem with forecasting in economics is surely that others are also forecasting and attempting to manipulate the system. With weather forecasting the result doesn't change just because it can be calculated.

matt4077 10y ago
It's not so much "trying to manipulate" (in the malicious sense) but that economic forecasts tend to be self-fulfilling prophecies. If you think a stock is undervalued you'll buy it , which will raise the price. If you think the economy will grow, you'll invest, which will cause the economy to grow. The result is the "efficient market hypothesis" which posits that all public information is already reflected in market prices to the best of our abilities (with a bit of swarm intelligence and darwinian selection making it unlikely that an individual can beat the market on ability). All that's usually applied to prices but other indicators like unemployment may simply correlate to such a degree that it extends to them as well.
- smallnamespace 10y ago
  It's usually applied to prices because the assumption is that the act of trading on perceived incorrect prices causes the price to converge to the market's expectations. Whereas for employment, you would need a mechanism that allows for 1) a public prediction of what the future unemployment path is and 2) a way for people to put money down if they disagree, such that it actually directly impacts unemployment.
- Normal_gaussian 10y ago
  I didn't mean manipulate in a malicious sense, but that doesn't really matter. You have a good point in that economic forecasts tend to be self-fulfilling, and that it is unlikely an individual can beat the market just with economic plays. Of course players don't just have traditional economic moves at their disposal, political power translates pretty well into economic manipulation - and that isn't just effective at a large scale.

nl 10y ago

I don't get it. Of course the economy is non-deterministic. That doesn't mean that useful forecasts can't be made. Statistical models are used because of this. I agree of course that deriving and presenting a single forecast from a statistical model is a dangerous process. Perhaps this was the link the author was portraying by their quest for an Isobar map rather than just "Rain/Non rain". (Also, I see others are looking at "economic forecasting" as price forecasting. I read this more as macro-economic forecasting, in which the outcomes are much to manipulate)

neximo64 10y ago
The problem is you can make forecasts but they're just guesses at best. The basic model for forecasts for example underestimates the occurrence of black swan events, plain and simple. That's why these days they use the stress test approach, assume the worst market conditions then see if a bank can cope.
- nl 10y ago
  The problem is you can make forecasts but they're just guesses at best. This isn't true. The current US unemployment rate is 4.7%[1]. It's pretty easy to look at the time series data, and come up with a likely range for the next reading. The possibility of black swan events (as well as inadequacies of models, measurement inaccuracies etc) mean this number isn't a sure thing: instead a statistical model provides a distribution of probabilities. For example, the probability distribution curve might give an 90% chance of the unemployment rate being between 4.5% and 4.9%. If we are trying to predict a year ahead, there much less certainty. Black swan events are just one of the causes of that, but it's still possible to forecast that the unemployment rate in December 2017 has a (say) 30% chance of being between 5.5% and 4.5%. This isn't guesses (well to be clear: I'm guessing above). You model the inputs, and build a Bayesian belief network (say) to model the interactions between events. Here's a version of The Fed's model: https://www.federalreserve.gov/econresdata/notes/feds-notes/2014/a-tool-for-macroeconomic-policy-analysis.html https://www.federalreserve.gov/econresdata/notes/feds-notes/.... That's why these days they use the stress test approach, assume the worst market conditions then see if a bank can cope. That's completely different to macro-economic forecasting. [1] https://data.bls.gov/timeseries/LNS14000000 https://data.bls.gov/timeseries/LNS14000000