Executive decision support for modern business management

Executive decision support for modern business management

Decision support systems are currently available for all medium and large companies around the world. Usually the scenarios extrapolate the central financial items over the planning horizon, frequently without explicit connections to the key managerial decision dimensions.

Acceptable values of financial ratios and holdings (return on investment, cash, etc.) are not automatically guaranteed. Instead, time-consuming adjustments of numerous financial items are needed to reach an acceptable base strategy. At the same time, the leeway for considering large sets of alternatives is reduced. An explicit connection between financial variables and the critical decision dimensions of the firm is scarcely implemented in practice, even less so a quantification of the corporate risk surface. Risk analysis is nothing new in medium and large firms, but it is mostly done through stochastic analysis of partial processes. The impact on the firm as a whole is seldom rigorously measured.

Illustration through a case study

To make my thoughts concrete I will walk through a practical case using only publically available financial information. I have not discussed the details of the financial information, the corporate plans or the parametric conditions with representatives of the firm in any stage of the process. I have made corresponding tests of the system on many different small, medium-sized and large firms in different branches: big public firms in forest, fuel, shipbuilding, machine and pharmaceutical industries, a Big 4 auditing firm, a medium sized firm offering accommodation solutions for marine and off-shore industries, a medium-sized firm mechanistically providing basic financial information, a small accounting consultancy firm.

The necessary model parameters are automatically calculated using available public financial information (table 1 in pdf file, all tables and figures are in the same file). The idea to this automation emerged in discussions with business analysts at Suomen Asiakastieto Group Ltd during 2017. We identify key variables in the parameter space, such as the corporate social media index (SMI), having a direct relevance for the market share. We cannot sell more than our share of the market. The idea to this index emerged in discussions with a risk assurance representative at PwC Finland in the fall of 2017.  Using these parameters, we calculate the corporate long-term basic strategy (table 2) that is confirmed by the management after due adjustments, for example of investments, inventory management, new issues and the cash dividend policy. The managerially confirmed long-term strategy serves as a guiding line for the corporate yearly budget and monthly allocations/follow-up (table 3).

In table 3 we find on an A4 the following key information:

  • The confirmed long-term strategy, in this case covering 5 years
  • The confirmed budget for the next year
  • The cumulative monthly follow-up from January to March in this example
  • The financial information for the realized quarter
  • The projection to the end of the financial year.

We summarize on this A4 the corporate profit and loss statement, assets & liabilities and cash flows. We define the cash flows as follows:

Initial cash + operating cash flows in compliance with IAS 7 ± financial items, cash dividends and investments = Terminal cash.

The idea to summarize the financial information on an A4 emerged in discussions with Chairman Riku Lehtinen at Strategic Accounting Finland Ltd, a company with more than 40 years of experience in strategic accounting consultancy and education.  Since the base year is 2016 in this example, it is convenient to compare the first strategy year (2017) with the realized numbers in order to identify a possible need for adjustments.

In addition, a nonprofessional can assimilate the information summarized on this A4 and form a better picture of the economic situation of the firm than what is possible through partial reports gathered from different parts of the organization.

For the follow-up to work perfectly, we need 22 items monthly from the accounting system (the general ledger) of the firm. This data input is concentrated on a separate page (table 4) to keep the definitions on table 3 untampered.

A paradigm shift in corporate executive decision support

The presented system and especially the summarizing A4-information in table 3 represent a paradigm shift to modern  "Executive decision support", where planning and budgeting are made more efficient and the coordination in the management group is improved.

At first sight, the presented model does not deviate significantly from the systems currently operational in large firms. However, the built-in components of the system provide possibilities to more diverse risk modelling and the coverage of larger sets of alternatives than what is possible through traditional risk analysis. Furthermore, the artificial intelligence based HPC-platform GHA (Genetic Hybrid Algorithm ) allows unique possibilities to connecting holistic firm models with corporate technology involving challenging computational problem solving.

Modern risk analysis

In figure 1, I present a risk analysis of the case corporation, where I have used massively parallel processing of the parameters in the firm model. The risk analysis comprises three stages:

(i) Determining unique combinations of the central parameters of the firm
(ii) Solving the firm model for each parametric combination
(iii) Determining the downside-risk of each solution.

Figures 1-2 require solving thousands of multi-period strategy alternatives, thousands of stochastic simulations for each strategy and the calculation of consistent multi-period financial statements for each strategic combination. As a result, in this example 102.4 billion complete internally consistent multi-period financial statements are computed. The system has undergone thorough consistency checking and control of the heap memory usage (cf. Valgrind-check). The figures – with 4096 points in each – depict the risk surface of the firm in the relevant parameter space. The green surface indicates the level below which the discounted total profit of the firm falls with a probability of at most 2.5%. The blue surface indicates the expected profit at different parameter levels. Each point on the surfaces represent a strategy for which consistent financial statements can be produced exactly. Figures 2-3 demonstrate that the system is operational with arbitrary planning horizons, in this case 100 years: the mathematical task stays computationally tractable when increasing the planning horizon.

With a long-term calculation, we can determine the stability of the managerial plan. Figures 1-2 would absorb 3 and 66 days of continuous computing with one computer having the clock rate of one Cray XC40 processor. When using the parallel processing modules of GHA, the computations are completed in 1 and 23 CPU-minutes respectively.

The computational platform GHA

The GHA-library used in the above presentation  supports problem solving over a wide range of techniques covering mathematical programming methodology, high-performance simulations, artificial intelligence based techniques or a combination of these. The library is written in strict ANSI C with support libraries in C++ and F90. Connectability to R through a C++ interface. Projects to be written in, e.g. C, C++, F, Open source R or Matlab (Octave). The library contains MPI-wrappers for parallel processing on supercomputers. A complete replica of GHA is copied to each parallel machine for maximum intelligence at the processor level. Different solution strategies can be applied simultaneously in different processor clusters. Using Early Mesh Interrupt allows immediate termination when the computational problem is solved somewhere in the mesh. Any external algorithm can be connected through the accelerator. Conversely, GHA can be integrated in other systems. Thanks to the accelerator function the researcher can pose the important question: how can I solve the current numerical research problem by invoking the best computational tools available? The system has been extensively tested and reported in scientific journals in finance, engineering, applied mathematics, statistics, operational research and physics. In particular, GHA has been developed and used in the following areas:

1.    Scientific computing

  • Difficult (box-constrained) global optimization problems.
  • Challenging problems in computational physics.
  • Constrained optimization: non-linear (disjunctive) MINLP-problems
  • Vector-valued time series modelling
  • Simulation
  • Cross-disciplinary HPC-problems
  • Integrating forecasting and optimization in portfolio management

2.  Firm modelling
3. Recursive portfolio modelling using GHA (banks, investment companies, financial expert organizations).

The system has proved to be an ideal flexible tool for industrial/academic research and development.

GHA is the academic life achievement of the author.

We are currently seeking partners for academic/commercial cooperation with Åbo Akademi University.

Acknowledgement: I am grateful for the advice and support of the experts at CSC IT center for Science achieved over many years.

Image: Thinkstock


 

The Author is Professor of Accounting and Optimization Systems at Åbo Akademi, School of Business Economics.



References:

Östermark, R (2018): GHA: A generic artificial intelligence based HPC platform for cross-disciplinary R & D. (PDF presentation)

Östermark, R (1999): Solving irregular econometric and mathematical optimization problems with a genetic hybrid algorithm. Computational Economics 13:2 pp. 103-115.

Östermark, R (2015): A parallel algorithm for optimizing the capital structure contingent on maximum value at risk. Kybernetes. The International Journal of Systems and Cybernetics, 44 No 3, 384-405. (http://www.emeraldinsight.com/doi/abs/10.1108/K-08-2014-0171).

Lahti A, Östermark R, Kokko K: Optimizing atomic structures through geno-mathematical programming. Forthcoming in Communications in Computational Physics.

 

Published originally 20.06.2018.

More about this topic » Go to insights and news »

Ralf Östermark