TROLL is an integrated software system for econometric modelling and statistical analysis. TROLL is the primary econometric modelling and analysis tool at central banks, statistical institutes, ministries of economy and finance, private companies and research institutes across the world.

TROLL offers a complete environment specialised in processing of analytics, including data storage and manipulation, model editing, regression tasks, model simulation, analysis of systems of equations, report design and publication, and high-level customisable interaction with spreadsheets and RDBMS.

TROLL is a powerful and modern tool that can operate on PC and UNIX machines (or simultaneously on both); It is used in central banks, banking and finance industry leaders, ministries of finance and economics, agriculture, economic affairs, planning, central statistical offices, international organisations, universities and public research institutes.

Send your questions about TROLL to troll@hendyplan.com

TROLL is fully portable on most PC and UNIX environments:

• all MS-WINDOWS versions beyond NT;
• LINUX;
• IBM-AIX, SUN-Solaris, SUN/OS, HP-UX.

TROLL offers the entire set of objects needed to enable the development and fine-tuning of econometric and modelling applications.

• The DATA object is the most versatile one with numeric, strings, dates, boolean and Not Available, TYPEs. You may organise a DATA object as a scalar, an array with any number of space dimensions (such as a matrix), scalar time series, or array time series. All functions are available to address whichever slice of hypercubes. Econometric development becomes easy as all the tools are at hand from a TROLL interface.
• The MODEL object is as unique: your system of equations are arranged into libraries that can be retrieved at any time, combined, merged into any direction, ... TROLL has all the necessary tools to make them understandable from an analytical point of view.
• The DATE concept offers every type of periodicity, from high-frequency data attached to calendars to daily data: annual, quarterly, monthly, weekly, daily and business daily, hours, minutes, seconds. User-defined periodicities add further flexibility.

This software has unrestricted possibilities: DATA can take on any number of values, object names can be of any length, models can have any number of equations and the number of DATA objects in a database is unlimited.

TROLL uses a powerful modelling language that is designed for large systems. Beyond the standard algebraic notation, operators are available for the computation of differences of terms over time periods (i.e. an expression at date t minus the same expression at date t-1); conditional expressions,  sum or product of any term over space indexes incrementation; functions such as MAX, MIN, trigonometric ones, etc.

From a structural point of view, TROLL is able to automatically re-order a system of equations, by looking at all dependencies and treating equations with a non-normalized left hand side, i.e. written as term_left_hand_side = term_right_hand_side, or term_left_hand_side = 0, or 0 = term_right_hand_side. The library also comprises functions for the manipulation of texts for equations, offering unlimited capabilities in developing your applications. The identification of variables and terms in equations, for instance, is made extremely easy. So is the extraction of symbolic derivatives in any term, at any order, for further model structure analysis.

TROLL offers all the common econometric methods for professional time series' analysis and forecasting:

• Linear single equation regression methods, with Ordinary Least Square, generalized least squares for first- and second-order autoregressive with correction procedures such as Cochrane-Orcutt and Hildreth-Lu, Two-Stage Least Squares, with distributed lags and polynomial distributed lags computation options;
• Non-linear single equation regression methods, using Non-linear Least Squares;
• Estimators for a system of non-linear equations on its coefficients with techniques such as Full Information Maximum Likelihood, Three-Stage Least Squares, Two-Stage Least Squares and Zellner's "seemingly unrelated regressions";
• VAR and VECM models using the Johansen co-integration algorithm for the latter one.

The integration with the MODEL object is especially interesting as it is possible to invoke this or that method from its explicit form i.e. the algebraic notation instead of pre-defined terms that are already normalised. TROLL distinguishes between symbols of linear versus non-linear forms and applies the right method accordingly.

Consequently, all tests and statistics are also possible. Durbin-Watson, F-Statistic, Log-Likelihood, normality, heterskedasticity, etc. All tests, every result and any type of diagnosis not only comes out as an output for a regressions' display report but also as individual DATA that can be post-computed.

TROLL offers state-of-the-art techniques for model simulation and analysis. TROLL solver is known for being extremely powerful and accurate in solving non-linear systems. Its programs are optimized in both the numerical analysis and IT-wise: there is no need to re-order equations, no need to normalize terms, no limitations in the number of equations or complexity. Common 256 Mb RAM 2gHz PCs can solve non-linear backward models in a few minutes for large sets of concurrent blocks of 5,000 to 10,000 equations.

The TROLL simulator comprises:

• Static and dynamic simulation modes, and all gradations of modes between these two, static every n periods and dynamic otherwise;
• Algorithms to solve many different models, NEWTON and relaxation methods (GAUSS-SEIDEL, JACOBI);
• Ability to compute many different models, backward and forward-looking models with state of art techniques like FAIR-TAYLOR, Statck Time;
• A host of parameters for simulation tuning, convergence/divergence criteria, damping factors, simulation initialisation in the first period, jacobian matrix re-evalation, condition number in matrix inversion, matrix sparsity, residuals on variables at every iteration, etc;
• Ability to automatically linearize a model for any period, computing of eigenvalues and vectors, jacobian matrix symbolically as  well as analytically;
• etc.

A variety of tools can be linked to these simulation techniques for the tracking of results as iterations are progressing, comparisons of results in a pool of shocks, etc.

The TROLL database interface is standardised. Any external file format can be used and processed by TROLL's kernel. By importing or exporting data or linking a database, the content becomes available straight away for further processing. These interfaces are dynamic vs. external formats.

So far many formats are available:

• one open format making external files of TROLL DATA exchangeable and compatible with other software;
• text formats compatible with major brands such as AREMOS (c), TSP(c),
• TROLL binary formats, including one which is compatible between UNIX and MS-WINDOWS,
• FAME format, making TROLL and FAME two products easily connectable,
• virtual databases in memory for high speed computation.

The list is not exhaustive as it is possible to plug any format for any types of specifications or needs.

TROLL hosts its own library for tables and graphs display. A wide range of details are possible thanks to a windowed, menu-driven, pool of functions for a programming-oriented approach. All standards are available:

• all styles of curves, i.e. line, bar, pie, etc.
• 2D/3D effects,
• control of areas, i.e. title, legend, areas,
• tuning of presentation, i.e. fonts, thickness of curve, colours, etc.

Other reporting facilities are available in TROLL Addons, see Advanced Reporting System, taking advantage of the powerful TROLL to MS-EXCEL interface.

The TROLL Programming Language (TPL) is complete enough to allow development of applications of any size. As it follows international normalisation standards it is fast to learn, both for economists, enabling them to develop their ideas and experimentations, and for IT-profiles, providing end-user applications in areas of statistics and econometrics.

TPL is a 4th generation language, meaning it brings to developers a way to merge in the same code, programming statements and TROLL functions or commands. As it is a compiled language (TROLL hosts its own TPL compiler), the result is very fast routines, inheriting the power of the TROLL kernel. In many cases a TPL program is faster than its equivalent C-routine because of usage of the TROLL built-in functions, with a code 5 to 10 times smaller: do you see the improvement in productivity thanks to such an environment?