What is financial risk

This article explains what is financial risk and how Aidvisors control it.

Why the 3D risk response is important

Below is an interactive 3D chart showing the risk response of Genesis Risk Management applied to the S&P 500 index.


The chart above shows tens of thousands of investment strategies with the following characteristics:

  • X axis – Financial Risk: the volatility measures both financial gain and financial loss
  • Y axis – Risk of Financial Loss: the negative volatility measures financial loss only
  • Z axis – Annual Return: the compound annual growth rate measures financial gain only
  • Color – Sortino Ratio: the Sortino Ratio measures the balance between financial gain and financial loss

The question we want to answer with the 3D risk response is:

How efficiently does the AI remove, reduce or transfer risk of financial loss?

The short answer

The remainder of this article provides the long answer. But if you don’t have time, here is the short answer:

If the 3D risk response looks like a cloud with no direction, chaotic and disordered, that means the AI could not understand the logic behind the financial risk of that asset.

This is why we provide the 3D risk response chart in every AI Report. It allows to find out if the machine learning pipeline could understand the logic to control the financial risk related to that asset.

There are a few machine learning pipelines which are tagged as dysfunctional. We clearly identify such dysfunctional AI Reports with an important notice at the top when it occurs and we don’t recommend to use those. Even if the annual return is higher, the probability is too high that it was out of mere luck. Read more about this in the disclaimer.

You may stop to read here if you don’t have time for the long answer.

The long answer

Transparency is important to us. This is why we take time to explain the AI stuff.

A flock of birds

When my kid look at the 3D risk response, he sees a bird flock.

A bird flock in a compound-V formation

Most of the time, the 3D risk response looks like a compound-V formation. The mechanics behind such a bird flock formation are not fully understood yet. But there are some well-documented bird flock properties:

  1. Each bird acts out of simple self-interest
  2. All birds share several characteristics common to their species
  3. Birds are exposed to similar environmental conditions such as gravity and turbulent air
  4. Some birds are clearly leading other birds, which means they are better at facing environmental conditions because of their individual characteristics

One can draw a parallel with the investment strategies characteristics:

  1. Each investment strategy acts out of simple self-interest (reducing risk of financial loss)
  2. They share common characteristics and tactics to manage risk (specific to the architecture of the machine learning pipeline)
  3. They are exposed to the same financial risk conditions (such as specific levels of stochasticity and heteroskedasticity)
  4. Some investment strategies have better annual return than others, which means they are better at facing the financial risk conditions because of their individual characteristics (hyperparameters)

The leading birds

In a bird flock, the leading bird will be replaced eventually by another leading bird because of fatigue or changes in environmental conditions such as wind direction. By identifying the common characteristics of the leading group, we increase the probability to pick one of the leading investment strategies at any time.

The leading birds in a V formation

There is something reassuring about the leading investment strategies at the tip of the triangle. Such clear convergence around a specific point (the tip of the triangle) is only possible because of common limitations. The leading group is very well defined by some measurable characteristics. Without changing the machine learning pipeline design (similar to bird species) or without changing the financial asset (the environmental conditions) there is little room for changes in the characteristics of the leading group.

What our system does is to identify that group of leading investment strategies and to continuously assess the current financial risk and provide almost the best possible risk response. The result is an higher annual return compared to doing nothing (buy & hold).

Risk response limitations

  • Is it possible to avoid financial loss? No
  • Do we have to accept to lose money to potentially make money in the long-term? Yes
  • Is it guaranteed that a financial asset will provide positive return on investment? No

How unfortunate is that? As a kid we believed in magic. At some point we stopped to believe in magic because we were exposed to tangible limitations, whether it was our own limitations or our environment limitations. So there is no magic here at all. It’s all about numbers, probability and computation power. That’s the machine learning promise after all, isn’t it?

So let’s explore those risk response limitations a little bit.

We have to accept financial risk to make return on investment.

In the chart below, we can see the investment strategies are getting above 15% annual return at around 12.5% volatility. Without financial risk, which includes financial gain, it would be impossible to have a positive annual return at all.


It is NOT possible to remove financial loss completely.

In the chart below, we can see the investment strategies are getting above 15% annual return at around 7% negative volatility. Trying to reduce financial loss under that level can only reduce financial gain and annual return.


The machine cannot control unknown unknowns.

Risk can be divided in two: there are known unknowns and there are unknown unknowns. The latter are the unidentified risks and they are made of both negative risks (threats) and positive risks (opportunities). Trying to reduce risk at a level where we would control even the unknown unknowns, means we would lose opportunities that we could not identify in advance.

Consequently, the risk response is limited in its ability to reduce financial loss without reducing financial gain. Below we can see on the right that it’s possible to reduce financial loss at a rate higher than financial gain. But on the left of the inflection point, around 12.5% volatility, we reduce financial loss at a rate lower than financial gain.


The chart above shows what we call the efficient frontier of the risk response. While it has a different shape compared to the modern portfolio theory, the concept and measurement is the same. The shape difference comes from the fact that here we measure a portfolio of investment strategies on one financial asset, not a portfolio of financial assets with no investment strategy.

Bottom line:

  • We have to accept financial risk to make return on investment.
  • It is NOT possible to remove financial loss completely.
  • The machine cannot control unknown unknowns.

Those limitations are due to what is called heteroskedasticity. But you don’t need to understand that fancy term. You just need to understand there are tangible limitations in the ability of an investment strategy to reduce financial loss.

The investor belief VS the machine learning purpose

As an investor you have to make a decision. You have to decide to invest in a financial asset you believe in. A belief is an acceptance that a statement is true. You have to believe there will be a positive return on investment in the long-term on that asset.

The machine learning pipelines we offer just assume that your belief is the right one. Assuming a positive return on investment in the long-term, the machine learning pipeline aims to help you reduce financial loss. But you had to accept financial loss from the start when you decided to invest.

The AI Reports allow to find out whether a machine learning pipeline is good because of its learned skills or because of mere luck. Differentiating the good from the bad machine learning pipelines is important. It cannot be good if it doesn’t understand the logic to control the financial risk. Because we are transparent we already identify the bad ones for you. But still, you will develop your own preferences and you are the ones who know your needs or the needs of your customers.

Read the AI Reports to find out if the risk response matches your desired risk tolerance.

Take time to read the dysfunctional AI Reports to learn how to differentiate bad 3D risk responses from good 3D risk responses.

We leave you with another pretty good 3D risk response. Here we use the results of Genesis Risk Management applied to the gold price.


For the AI professionals out there

The 3D risk response is generated when executing hyperparameter optimization of one machine learning pipeline. The worst investment strategies, with lowest annual return, are the results of random misconfigurations of the pipeline. Those misconfigurations would never happen in production since we know they are incorrect by design, but we allowed the system to generate such misconfigurations for exploratory research.

It would be a mistake to believe that the investment strategies shown in the 3D risk response are simple trading algorithms. That would make Aidvisors as simple as algorithmic trading and it is NOT the case. Actually, without going in much details, each investment strategy that you see in the 3D risk response is a huge ensemble. Each investment strategy is made-up of hundreds of thousands of trading algorithms, which are optimized by thousands of predictors such as neural nets, which are optimized into ensembles of predictors. For each investment strategy, there are a minimum of three (3) evolution chains which track the best trading algorithms, the best predictors and the best ensemble of predictors. For instance, you could visualize each evolution chain as a movie of daily 3D risk response, each frame of the movie being the 3D risk response for one day. And there are three (3) of these movies in parallel.

At some point in the future, we may take the time to create a beautiful interactive movie of all those moving parts behind each investment strategy. But only for our best customers who suffer from intellectual curiosity. Don’t hesitate to let us know if you would be interested in such a movie, it could be a beautiful interactive piece of art to display in your living room!

Otherwise, keep in mind that the 3D risk response is actually a very simplified view of what is happening behind the scene. The 3D risk response is just showing the results of a few hyperparameters used to fine-tune how the machine is learning. Actually those hyperparameters define how to optimize and learn from a lot more other parameters: the parameters of the trading algorithms, the parameters of the predictors, the parameters of the ensembles, etc. And the optimization of the latter parameters occurs on a daily, weekly or monthly basis and are tracked in huge data structure which we call evolution chains.

The most important to understand is:

Aidvisors are not trading algorithms. Aidvisors are complex machine learning pipelines which learn every day and adapt to changes.

Next article

Popular AI Reports