FAQ: How to work with AI Reports

An AI Report is a document describing the financial performance of Aidvisors (a.k.a. Artificially Intelligent Advisors) that have been trained on one or more financial asset(s).

The AI Report’s purpose is to inform financial professionals and advanced investors on the Aidvisors risk response throughout the years in different market cycles. It allows to assess whether or not the Aidvisors are fulfilling the needs identified in term of risk management. Better annual return are often observed and they are always welcomed, but they are just a possible side-effect of a better risk management strategy.

The report usually contains the following sections:

• Risk Management explains how Aidvisors control financial risk
• Risk-Return Optimization explains how Aidvisors optimize annual return
• Machine Learning explains how Aidvisors are trained to make the best decisions
• Decision Making provides details about preventive actions that are suggested

Algorithmic trading is about one deterministic set of instructions, a unique recipe. The problem with recipe is that it doesn’t adapt to changes, it’s based on the assumption that what happened in the past will repeat itself very similarly. But repetition is not the rule with financial markets.

Aidvisors (a.k.a. Artificially Intelligent Advisors) are based on ensemble learning and evolutionary optimization. In other words, Aidvisors are a mix of many recipes (instead of one unique recipe) and the recipes are continuously improving (instead of one deterministic recipe).

Aidvisors are big ensembles of trading algorithms that are continuously adapted to the changing market cycles. For instance, with the machine learning pipeline called Genesis Risk Management, hundreds of thousands of trading algorithms are optimized on a daily basis, and they are optimized using thousands of predictors, such as neural nets, that are trained on a monthly basis. And those figures are just the minimal figures.

The main reason to use advanced machine learning models is because financial markets are stochastic and heteroskedastic, which means that they are nearly random. Without adapting to changes it won’t manage the risk correctly and won’t stand a chance to outperform the market in the long-term.

As a French statesman once said: “The great recipe for success is to work, and always work.” Our solution is to put AI at work to manage financial risk. Our recipe is not a single trading algorithm. Our recipe is a whole AI system doing intensive work to boost numerous trading algorithms every single day.

The return on investment is the easy part. No wonder why you probably want to jump to the hypothetical growth chart to visualize the return on investment. At the end, investors are really just interested to know how much value was added to their account, right? Wrong!

No need to say, there’s a difference if you are a multi-millionaire because you won the lottery compared to having started a successful business that made you multi-millionaire. In the first case its the result of luck. In the second case its the result of a lot of work and hardship. If you only care about the money and not how to get there, farewell and good luck! However, if you started a successful business once or twice, there’s a high probability that the skills you learned will help you having success in the future. The key to repeatable success is what you learned.

What you learned during your past experiences is the most important to shape a successful future. But past results are no guarantee of future performance. Consequently, when you read an AI Report you should be more interested in what the Aidvisors (a.k.a. Artificially Intelligent Advisors) have learned than the result they would have got.

By focusing on risk management, it allows to identify what the Aidvisors have learned. The value of Aidvisors is what they learned, not just the return on investment they would have got. The return on investment is important as a measurement of how the Aidvisors applied what they learned. So the return on investment is important as a secondary topic.

Take time to understand the whole structure of the AI Report. Depending on the needs identified, some sections may be more important than others.

Often, you’ll want to jump to the hypothetical growth chart to visualize the return on investment. But that could be a mistake, because past results are no guarantee of future performance.

After taking time to analyze some of the other interactive charts available, you’ll start to see details in the risk management strategy that may better respond to the needs identified and desired risk tolerance. Spotting these details will be key when you start to compare with other AI Reports.

Take time to look at many AI Reports. It’s only by comparing many AI Reports that you can grasp the diversity of risk response available. Aidvisors are all about diversification. It’s about diversifying the financial assets to invest in and diversifying the investment strategies to invest with. By diversifying you limit exposure to financial risk.

On our side we work on the next generations of AI. More AI Reports will become available so you can compare with the previous generations AI Reports and verify whether the enhancements better respond to the needs identified.

Only when you conclude that an AI Report fulfills the identified needs and desired risk tolerance should you subscribe to the daily notifications so you can stay informed on the suggested actions. Never use the AI Report’s preventive actions available in the Decision Making section since it can be outdated. Always use the daily notifications to take investment decisions.

The AI Report is the result of a very complex backtesting which takes days to perform. The complexity comes from the machine learning pipeline behind. Each step of the pipeline complies with many quality assurance principles.

Here are some of the principles we apply to ensure accuracy of the data presented in the AI Report:

• Time-Sensitivity: The many machine learning models are trained only with past data, never with future data. For instance, if a neural net was used to make a decision on a given day, this neural net was trained with data available at the end of the previous day, as if the neural net training occurred during the evening and night after the market was closed. This principle ensures the machine learning pipeline doesn’t suffer from future bias. It would be too easy for the machine to perform well if it would have access to data from the future.

• Data-Immutability: Machine learning models not only consume data but also create new data, which is consumed by other machine learning models down the stream. When new data is stored in database, including when machine learning models are stored, it is never modified ever after. This principle ensures re-execution of the machine learning pipeline will produce repeatable results each time. Without repeatability their would be no way to verify accuracy.

• Data-Comparability: Machine learning models could not apply what they learned on new data if it would not be comparable to the old data they were trained with. Without data-comparability, machine learning models would become obsolete over time, not because the situation changed, but because they cannot even understand the new data. This principle ensures durability of the machine learning pipeline. Durability is key for constant financial performance. Hence, all financial data is processed in a way that normalizes the features used for machine learning.

No! The AI Report doesn’t replace financial planning. This is one reason why the AI Report targets financial professionals and advanced investors. We are planning to offer Aidvisors (a.k.a. Artificially Intelligent Advisors) that could assist with accounting and legal needs. But this service is not ready yet.

Due diligence is required in addition to the investment strategy described in the AI Report:

• The AI Report cannot take into account the specifics of the investor’s situation and type of account used. For instance, tax benefits and liabilities may vary a lot depending on the tax treatment that applies. It is important to do due diligence on how the tax treatment specific to the investor financial situation impacts cumulative returns.
• Within the AI Report it is assumed that dividends are re-invested and that transaction cost is minimal in comparison to the amount invested. The machine learning pipelines are configured to minimize the number of transactions and to favor long-term investment. But even if it is never reaching day-trading transaction rates, the number of transaction may increase significantly during period of high risk. It is important to do due diligence on how transaction fees and dividends impact cumulative returns, including tax considerations associated to transactions and dividends.

Let’s say you found an AI Report that fulfills the identified needs and desired risk tolerance that were identified during financial planning. It’s time to subscribe to the daily notifications so you can stay informed on the preventive actions the Aidvisors suggest. Never use the AI Report’s preventive actions available in the Decision Making section since it can be outdated. Always use the daily notifications to take investment decisions.

Daily notifications are usually delivered 30 minutes before the exchanges close, normally around 15h30 EST (Eastern Standard Time) for North America. This allow to make decision based on suggested actions before the exchanges close. And the good news is that most of the time the transaction could complete the next trading day without much impact (see next question for more on that).

Daily notifications are available through private messaging and secured API to a limited number of users. Refer to the Getting Started article for more information.

You may be afraid of discrepancy between the AI Report and reality because the stock price is different or because the transaction could not even complete on the same day. This is why we provide the Cost of Failure as a measurement in the Decision Making section of the AI Report.

The Cost of Failure is measured with a simulation of the worst case scenario for which all transactions would complete at the end of the next trading day. Most of the time, you can verify in the AI Report that such discrepancy would not cost more than 0.1% annual return per failed transaction. The Cost of Failure is usually insignificant in comparison to the annual return.

The machine learning pipelines are configured to minimize the number of transactions and to favor long-term investment. So it’s never reaching day-trading transaction rates for which discrepancy quickly cumulates into a snowball effect. Because the number of transaction are limited, the Cost of Failure is usually low.

Take a look at the next generations of AI we work on. Maybe you can find a better fit in upcoming enhancements.

We can also build custom machine learning pipelines to fulfill more specific needs. You can get exclusive access to the new Aidvisors (a.k.a. Artificially Intelligent Advisors) we build and configure for you. Or even run your own proprietary on-premise Aidvisors. Refer to the Getting Started article for more information.