The Predictive Content of Time Series Dr. Manfred Hrter www.albconsult.de

1
The Predictive Content of Time SeriesDr.
Manfred Härterwww.albconsult.de
2
Preliminary Remarks
Our main interest is in forecasting. Our
clients expect from us better forecasts than
their competitors do have. Our approach is an
optimistic one. We believe that there is some
stability in the systems we are interested in. To
detect weak signals for turning-points is our
main task. The traditional techniques as e.g.
Moving Averages (and other smoothing procedures)
as well as Futures (where available) cannot
help. Especially Futures - quite opposite how the
term sounds - do merely reflect expectations
already contained in spot market prices.
3
Brent Spot Price and Futures
4
Introductory Remarks

• The main problem for forecasters in practice is
  anticipating turning-points.
• The common techniques of Technical Analysis in
  form of trend lines and MACD may improve
  forecasting results.
• Especially the identification of resistance and
  supporting lines can improve forecasting results
  significantly.
• An objective method for identifying such
  resistance lines for time series can be get from
  frequency distributions.
• In the case of marked multi-modality we propose
  the idea of attractors as the basis for our
  forecasts.

5
The Case of Metals and Steel
6
HMS Time Series 1990 - 1998 Frequency
Distribution
7
Time Series HMS and MA (HMS,5)
8
HMS

• There are three different price regimes in the
  90s.
• Prices dont change or if they change they do it
  significantly and fast.
• If you can rely on that your forecasts will
  improve.
• The hardest test for that has been the price
  crash at the end of 98.
• In our interpretation it is quite clear that from
  the very beginning there is to expect a direct
  shift from price attractor 3 (130) to price
  attractor 1(80).
• The price movement was very sharp from the very
  beginning and if you have such a frequency
  distribution you can expect such a plunge.

9
The Non-Random Walk of Oil Prices

• Originally we developped our concept with regard
  to oil prices.
• That frequency distribution shows a marked
  bi-modality over the last twenty years.
• The astonishing observation has been that most
  oil price forecasts have been in the mid between
  the two price regimes, i.e. the price with the
  statistically lowest probability.
• Interpreting the two halves of the frequency
  distribution as two different parts was the
  analytical basis for correct forecasts in the
  last 12 months.
• Especially with regard to the April forecasts it
  was the cornerstone for predicting the
  turning-point of oil prices correctly.

10
Chart-Technical Interpretationweekly basis
(February 2000)
USD/b
11
Resistance and Supporting Lines- upper price
regime
12
Turning-Point Signal for Brent BlendApril 2000
USD/b
13
Brent Short-term downward trend broken
USD/b
Fächer-FormationAbwärtstrend schwächt sich ab
14
Resistance Lines dividing two price regimes
15
Oil Price Forecast PROGNOS December 1999
16
The Case of Currencies

• A quite unusual task is forecasting the future
  price of a new - synthetic - currency like the
  euro.
• As an analytic tool we used again the frequency
  distribution, additionally to the normal trend
  line approach.
• By doing so, in retrospect it looks very easy to
  forecast the continous depreciation of the
  exterior value of the euro, quite opposite to the
  dominating opinion of the market makers.
• Even the final turning-point of this negative
  development was principally predictable - far
  away from what the market originally expected.

17
Time Series US/Euro
18
Frequency Distribution Exchange Rates US/Euro
19
ActualExchange Rate Forecasts
20
Summary

• We propose Technical Analysis as a necessary tool
  at least for controlling the results of more
  sophisticated modeling.
• The identification of at least temporarily stable
  trends improves forecasting results
  significantly.
• With regard to cases of a marked multi-modality
  we can derive such stable lines even from
  frequency distribution.
• We tend to interpret the modes of frequency
  distribution as attractors.
• From the multi-modal frequency distribution we
  have learnt that prices may develop in leaps.
• But astonishingly that must not deteriorate
  forecasting.