If we run a model against a hundred years of data, looking to predict the peak for the hundred and first year, the classical model takes the 100 peaks and applies a mathematical rule on how you model these peaks to predict the 101st peak.
The big drawback is the amount of data you waste and the fact that the model needs to be rebuilt if the threshold parameters are changed. This makes the process expensive and a drain on your resources. A better approach developed by 4-Xtra is one where the data isn’t wasted. You draw a high threshold, for example, 95% of the data; so, 5% of the data is very high and 95% is moderate (‘normal’). The moderate values are often called ‘bulk’ and the extreme values are peaks-over-threshold.
The advantage here is that we keep all the values in play, not just one, and model all these peaks.
Until now, the only mathematical approaches available for this have been limited to stationary data. The model created by 4-Xtra answers the question of how to modify this approach in a nonstationary manner to retain threshold stability. When creating statistical models there are three important parameters of the extreme value distribution: shape, scale, and location.
Shape is the most important parameter as it determines one of the three different classes of the distribution. These parameters are constant in the classical model where everything is stationary. Yet in the nonstationary case, the parameters are, in general, not constant anymore. This presents a huge problem — how to characterise the time dependence as there is no natural way to quantify an unknown.
The classical approach is to adopt some parametric dependencies and then estimate the parameters, but this is apparently both artificial and subjective, and leaves open the question of validation of the chosen model.
Within the 4-Xtra models, time dependence is incorporated by treating the shape, scale, and location as functions of various covariates of interest (which depend on time in their own right). This is a lot easier as you don’t have to think about the explicit functional time dependence, which simplifies things greatly.
This idea is not new, it has been known since Davison and Smith (1990). The key difference the team created is a new way to parameterise the model to achieve the so-called threshold stability (which was missing in old nonstationary models).
The comparison was stunning: when datasets were tested against a decades-old classical model and the new, threshold stable model, the latter was found to be about 10 times more accurate.
However, the more conceptual advantage of the 4-Xtra model is this: changing a threshold in a classical risk model requires the model to re-estimate all the parameters by rerunning the Markov Chain Monte Carlo simulation. This is a huge disadvantage for the classical regression implementation of statistical modelling. The 4-Xtra models have threshold stability, meaning that if the threshold needs to be changed (e.g., because we wish to test for another level of risk), then the parameters of the fitted model don’t need to be re-estimated and can merely be recalculated using simple algebraic formulas, making it significantly simpler for the users of the program.
Even more important is that being threshold stable makes the model ideal for a dynamic data feed. When the new data is collected and added to the dataset, the Markov Chain Monte Carlo (MCMC) simulator can use the previously obtained Bayesian solution as the new prior (rather than an uninformed prior as in the initial run of the algorithm), and this dramatically improves the efficiency of the model by up to 10 times.
This platform has been in development over the past ten years, and the 4-Xtra Technologies team has been working towards the ultimate goal, which is to provide a Software as a Service (SaaS) solution in the first instance to the financial services market, to allow banks and other financial institutions to benefit from their model’s ability to forecast extreme events across a number of use cases, such as portfolio optimisation, and trading of any number of asset classes, from securities to foreign exchange, and therefore enhance their risk management decision making processes and ultimately improve their sustainability and profitability. Using this platform gives institutions the ability to maximise all opportunities that are available to them.
With all these improvements and innovations, financial risk management and assessment is made accessible, accurate, and more cost-efficient. Risk management professionals and financial institutions can rely on 4-Xtra to offer a Software as a Service (SaaS) platform that they can integrate and manage as they deem fit. Aside from answering the need for an efficient and user-friendly model, the 4-Xtra team has also addressed the need for a reasonably priced platform.
In summary, 4-Xtra is a disruptive risk management technology, that is driven by the unstoppable force of regulatory change. Ultimately, the company’s goal is to challenge the traditional, to provide an innovative, efficient solution to address the needs of the financial services sector when it comes to risk management and risk modelling.
Those interested in learning more about 4-Xtra Technologies’ SaaS platform, or other services that may be made available to you, please contact the 4-Xtra Technologies team through email firstname.lastname@example.org. Our team will get in touch with you as soon as possible.