Use AI to conduct sensitivity analysis

Let's focus on the task of forecasting a company's net income. How can AI models look at the potential values of all the variables that go into the forecast and generate a probability distribution for the possible net income values? So we're talking about changing from a single value point estimate of net income, say $5 million, to a probability distribution such as this. There's a 65% probability that net income will be greater than 3.8 million, and just a 5% probability that net income will be less than 1 million. As an example, consider the height of men in the United States. The average height of a US man is five feet, nine inches tall. Our understanding of men's height is increased when we learn that only 10% of US men are less than five feet, five inches tall, and only 8% are taller than six feet one inch. Here are the steps that an AI model can go through to create a full probability distribution for forecasted financial statement values. As an illustration, let's focus on forecasting a net income probability distribution. First, AI can help us by identifying net income drivers. We need to identify the income statement line items that contribute to net income. Now, that's pretty easy. We just looked down the income statement. The sales amount impacts net income, as does cost of goods sold. Operating expenses, interest expense, and income tax expense also impact net income. Within the operating expense category, it is important to break out the total items as much as possible. For example, the factors that influence wage expense or insurance expense are much different from those that impact depreciation expense. Secondly, AI can help us develop probability distributions for each driver. This is where AI power and human insight combine to describe the uncertainty surrounding each of the identified net income drivers. For example, historical data can be used to determine how close the sales forecast point estimate has been in the past. This can then be used to specify, for example, the limits on the range within which the sales number is expected to fall. This is called the confidence interval. Third, AI can help us by quantifying correlations among drivers. Some of the net income driver items move together. For example, in years in which positive economic news causes sales to be higher than expected, that same positive news can cause optimistic landlords to increase the rents they charge more than the expected amount. Fourth, AI can help us through running simulations. With the net income drivers defined and their probability distribution specified, AI models can now run simulations of the possible outcomes. These simulations are kind of like games. In fact, one common simulation technique is called the Monte Carlo simulation after the famous gambling location. When my brother and I were boys, we ran simulations like this on sports teams. Our tools were pencil, paper, and dice to introduce the random element. Imagine what we could do now with AI models, whew. Fifth, AI can create the probability distributions. Using thousands of estimated net income amounts from the repeated simulations, AI models can now construct a probability distribution by answering questions such as these. What is the average net income forecast? What is the net income forecast amount that sits at the 95th percentile, meaning that it's greater than 95% of the estimated values? What is the net income forecast that sits at the 5th percentile, meaning that only 5% of the estimated values are lower? And sixth, AI can help us by updating our data continuously. The estimated distribution is not a one-and-done thing. An AI model can monitor changes in the economic variables, business news, and company announcements, and incorporate this new information into the probability estimations for the individual net income drivers. This can be done every day, or even every hour for particularly important estimates, such as the net income distribution estimate done in the days or hours before the company publicly announces as its actual net income. The up-to-date probability distribution aids analysts in determining when the earnings actually announced are better or worse than what would have been expected given the prevailing attitudes about the company's economic situation. Here is the ChatGPT summary of the ability of AI to give better net income forecast understanding through probability distributions. AI can make net income forecasting far more powerful by one, capturing uncertainty in each input, two, recognizing relationships among variables, and three, running simulations to produce a probabilistic view of net income, not just a single point guess. AI can help make seemingly difficult tasks a lot easier just by helping us be systematic.