FAQ
Please find below frequently asked questions that you may find useful
1. How does finalyz pSci work?
The finalyz pSci program delivers forecasts for a broad range of financial market data. It is a complex, self-learning software package that combines various basic prognostic concepts such as filters, pattern recognition processes and decision algorithms. These routines are embedded in the framework of a nonlinear programming algorithm. The algorithms on which finalyz pSci is based were developed especially for the dynamics of the financial markets. They extract the relevant information from incoming financial time-series and estimate their prolongation into the future.
2. What kind of input is required?
Daily time series of financial products (e.g. share prices) are the required input and the depth of the historical data must be at least 18 months. Additional daily or periodic information about the dynamics of the financial market and the economic environment can be included as auxiliary data (e.g. exchange rates, integral indices, economic figures). The only limitation is presented by the calculating capacity. By default, up to 10,000 time series are used to generate the forecast.
3. What kind of output can be generated?
The output offered by the program is a subset of the input time series determined by the customer, over a time horizon, also set by the customer, of between three days and three months.
4. Is there any flexibility?
The subsets of the forecasted time series and the forecast horizons can be freely chosen, however, only one forecast horizon is possible within a subset. If the customer is interested in two or more horizons for certain financial products, then it makes sense to define the same subsets for different horizons.
5. How reliable are the forecasts?
Forecasts are basically statistical statements that predict the likely course of a process. The reliability of forecasts is generally measured by their accuracy and stability. In real-time tests conducted over the course of several years, it was shown that the forecasts made for all products have a positive medium stable correlation with a correlation coefficient of 0.1, and the group with the greatest forecasts even have a correlation coefficient of 0.25 (on a scale of -1 [uncorrelated] to +1 [fully correlated]).
6. How long does a calculation take? How does the calculation time relate to the number of time series input?
A forecast with 10,000 product time series requires about 20 hours. The computation time increases in straight proportion to the number of products, but can be reduced with increased parallelization with regard to the number of nodes used.
7. What are the requirements for the customer?
The customer must have Internet access and software that can download the forecasted data from a customer account on a Finalyz AG server. The data can also be accessed via SunGard.
8. What else is offered?
In addition to forecasting, Finalyz AG can also develop software for calculating and analyzing financial and risk data, using both established standard methods and its own techniques designed solely for this purpose. Finalyz AG also offers newly developed, dynamic trading strategies which, among other things, allow the forecasts provided by finalyz pSci to be taken into optimal consideration.
9. Why doesn’t Finalyz AG use the program itself to actively manage a portfolio?
That is planned for 2010 in the form of a retail fund.
10. What is Finalyz AG’s marketing strategy for finalyz pSci?
The software is installed on the computers of Finalyz AG. Finalyz AG offers every customer access to these forecasts, potentially even at exclusive conditions. Sale of the software is not planned.
11. What is the best way for a customer to use finalyz pSci?
In principle, any customer can align his or her fund or, more generally speaking, financial investments with the forecasts delivered by finalyz pSci. This involves various concepts. Forecasts can be used to confirm or reconsider customers’ own ideas, or customers can access direct buy or sell recommendations via suitable trading strategies offered by Finalyz AG.
12. What are the costs of using finalyz pSci?
The use of finalyz pSci data is offered in three steps. Use is free for the first month; from the second to the sixth month (depending on terms and conditions), use is charged at cost; from the seventh month on, use of the data is charged on a royalty basis.
13. Are there upgrades?
Yes, a new version of finalyz pSci with improved optimization and therefore increased accuracy is in development. Forecasting software is also being developed for short-time use (up to 1ms reaction time) to support direct algo trading.
14. Which scientific models is finalyz pSci based on?
Roughly speaking, the entire concept is based on the Nakajima-Zwanzig theory (M. Schulz, Control Theory for Statistical Physics and Other Fields of Science, Springer 2005). Unfortunately, this theory is very abstract in its construction, so the mathematical objects it contains (projectors, operators) must be replaced by suitable numeric procedures. However, precisely this is at the company core of the finalyz pSci software.
15. Which tests were conducted?
There are several teal-time tests with at least 1,000 input and output time series each. The last test used about 5,000 time series and was constantly monitored by a major bank/asset manager (which will not be named here). The corresponding report can be viewed at http://www.finalyz.eu.
16. How does finalyz pCom work?
The finalyz pCom program provides forecasts and up-to-date information about hidden market information (spot price, storage costs, interest rate levels) based on the futures price of a defined product in the commodity market. The program uses dynamic filters based on established market models. According to these models, there are various versions of finalyz pCom which are defined by relevant parameters. These parameters contain a variable set of degrees of internal freedom of the model. The optimal estimation of all couplings between these degrees of freedom is realized with the help of specially developed filters.
17. What is the scientific basis for finalyz pCom?
The dynamics of commodity trading differ significantly from that of the financial markets. This is due to the fact that commodities are usually only available in a limited quantity because of finite hauling capacities, and, as is the case with agricultural products, are often only offered seasonally. In addition, these products are characterized by a relatively slow change in demand, short-term fluctuations in demand must be compensated for by costly storage, and long routes of transport must be accommodated due to increasing globalization. However, these factors make such markets more transparent and easier to model mathematically. Our software versions mainly differ in their degree of complexity. All versions are based on the Schwartz models, which were developed in 2000 and have since become the standard. It should be noted, however, that the versions of finalyz pCom use considerable generalizations of these models.
18. What is required as input?
The input is made up of the futures price time series based on various maturities (e.g. from one month to three years for crude oil).
19. What are the possible outputs?
The finalyz pCom software implemented in the computing center of Finalyz AG can deliver permanent forecasts about the expected development of futures prices, spot prices and storage costs, which are of essential importance to commodities traders with respect to making buying and selling decisions. The forecast horizon is between one week and six months, and in some cases even longer.
20. Which tests have been conducted?
Three different degrees of freedom (2-4) were back-tested on the price of oil (ICE Brent Futures) with an effective duration of three years as part of a dissertation at the University of Ulm (Germany). As the degree of freedom increased, the back tests delivered an increasingly positive correlation coefficient from 0.1 to 0.3 (on a scale of -1 [uncorrelated] to +1 [fully correlated]).
21. How long does a calculation take?
Calculations are relatively time-consuming. The calculation time increases with the number of degrees of freedom N according to the power law N4, and is about two days for a commodity with three degrees of freedom.
22. What requirements must be met?
The requirements are internet access and software that can download the forecasted data from a customer account on a Finalyz AG server.
23. How reliable are the forecasts?
In the back test conducted at the University of Ulm (Germany), correlation coefficients between 0.1 and 0.3 were found (on a scale of -1 [uncorrelated] to +1 [fully correlated]), confirming an extremely high degree of reliability (far higher than weather forecasts over the same time period).
24. What is Finalyz AG’s marketing strategy for finalyz pCom?
The software is installed on the computers of Finalyz AG. Finalyz AG offers every customer access to these forecasts. The sale of specific software at exclusive terms and conditions is also possible.
25. What are the costs of using finalyz pCom?
The use of finalyz pCom data is offered in three steps. Use is free for the first month; from the second to the sixth month (depending on terms and conditions), use is charged at cost; from the seventh month on, use of the data is charged on a royalty basis.