|
Forecasting Basics
The scientific method and mathematical model used by
Stock-Forecasting.com is based on the idea of harmonic structure analysis of
stock prices time series.
It is obvious that some trends of stock prices have a wave
(non-periodic oscillating) structure and can be represented as a combination of
the number of harmonics with unknown and changing frequencies and amplitudes.
The main problem is a choice of such elementary functions and their parameters
(for example, for Sin(õ) it is amplitude and frequency) that it was
possible to represent any initial time series. Accordingly, it is necessary to
provide an opportunity for the automatic fine tuning of the parameters of
elementary “time - flowing” functions during forecasting.
The ordinary scientific methods cannot be used for the
separation of the variable parameter harmonics of “trend + noise” signal. The
special adaptive technique used by Stock-Forecasting.com’s mathematical model
includes auto-regression representation of wave series and both: frequencies
and partial harmonics amplitudes identification.
The peculiarities of the method are the initial time-series
decomposition on the slow (trend) and fast (oscillatory) components with the
help of digital filters (our “know – how”). The special software based on a new
method of forecasting has been developed.
For each step the following calculations are performed:
- the initial time series trend (“stock vs. time” 200 – 400 days
back in companies historical data base) is separated into the slow
(trend) and fast (wave) components by means of digital filtering algorithms;
- the harmonic components of both trend and wave terms as well as
the first difference and number of efficient harmonics are determined;
- the short term (1-10 days ahead) forecast based on the estimated
harmonic mode is calculated.
The source data contains the daily information on share prices for more than
9000 stock quotes and 500 indicators traded on traditional stock markets. The
database is updated daily and includes the following: open price, close price,
minimum and maximum. It is necessary to take not less than 200 days of
historical data (200-400 days) for a daily time interval.
As a result you receive a table which contains predicted data
in a specified interval. There are six columns: prediction date, predicted
values of an open, close, low, and high price. Also each table contains on
“Average” column which is the predicted average price for the day: (Open +
Close + Low + High) / 4. Each row corresponds to one day a head.
Here is an example.
Forecasted data. IBM
| Date |
Open
|
Close
|
Low
|
High
|
Average
|
| 02.05.2004 + 1 |
99.78
|
99.92
|
99.92
|
100.90
|
100.13
|
| 02.05.2004 + 2 |
99.90
|
99.25
|
99.20
|
101.08
|
99.86
|
| 02.05.2004 + 3 |
100.51
|
99.05
|
98.17
|
100.54
|
99.57
|
| 02.05.2004 + 4 |
100.40
|
98.48
|
98.45
|
100.84
|
99.54
|
| 02.05.2004 + 5 |
101.90
|
97.80
|
97.02
|
101.90
|
99.66
|
| AHR Errors, % * |
3.09
|
3.47
|
2.57
|
2.16
|
2.82
|
*Average Historical Relative Error ( AHR Error ), is
calculated by comparing the predicted quotes (open, close, low, high, average)
for every day of trade, going back 50 days from the last day of trade, to the
same trading day actual historical quotes.
Description
Last business day of trade: 02.05.2004 February 05, 2004.
FORECAST:
| |
02.05.04+1:
February 06, (Friday)
|
02.05.04+2:
February 09 (Monday)
|
...
|
| Open |
99.78
|
99.90
|
...
|
| Close |
99.92
|
99.25
|
...
|
| Low |
99.92
|
99.20
|
...
|
| High |
100.90
|
101.08
|
...
|
| Average |
100.13
|
99.86
|
...
|
|