VICO's platform is an AI-powered decision superiority platform that allows users to instantly generate expert-level probabilistic forecasts, scenarios, simulations and risk assessments for macro, policy, and geopolitical events—helping investors test views, identify mispricing, and size risk with greater confidence. Backtesting shows equal or greater accuracy than experts or prediction markets.
The VICO platform does the following: • VICO forecasts the probability of events occurring. This is in the form of "What is the probability of X event during Y time period?" For example, what is the probability of a ceasefire in Ukraine in 2026? • VICO can forecast in a set time period, i.e. in 2026 or by August 31, as well as a rolling time period, i.e. in the next 90 days or next year. • VICO performs What if scenarios. This takes the form of asking: "What if Z occurs, then what is the probability of X occurring in Y time period." For example, what if NATO sends troops to Ukraine, then what is the probability of a ceasefire in 2026.
VICO extracts real-time news from 150,000+ data and news providers using a proprietary LLM-powered data engineering pipeline to generate structured event data. The extraction pipeline is accomplished through multi-tool orchestration and decomposes each source record (e.g. news article, research report, earnings transcript, correspondence), into a uniform event taxonomy comprised of atomic event components.
VICO generates and maintains 1 million+ proprietary data transformations on our current record volume of 1 million+ event records, which convert raw, unstructured open-source news data into structured event data represented in a uniform event taxonomy. This taxonomy decomposes events into atomic units, including actors, entities, actions, impact, and geospatial coordinates. VICO event data are also processed through an analytics pipeline, which generates proprietary analytics on report reliability, including article sentiment, political bias, disinformation, misinformation, author personal and narrative bias, and foreign political ideological bias. This structured, analyzed dataset is proprietary to VICO and only available to VICO subscribers.
No. VICO models are powered exclusively by contractually licensed, commercially available API providers or through fair use rights from public releases.
Our forecast system consists of a multi-agent workflow to generate probabilistic forecasts that encode expert judgement reflecting scientific forecasting principles. The workflow includes an LLM-retrieval agent, followed by a second AI agentic workflow which instantiates an expert forecaster agent. Our models then execute a series of sequential computational steps that combine probabilistic reasoning, Bayesian inference, and data validation models that are aggregated through an ensemble workflow. In producing a final output, our system also performs additional validity checks to ensure against hallucinations or other errors and to maximize forecast calibration to live signals. Note: the particular means by which VICO makes forecasts is VICO's deep intellectual property. These details are restricted to Company employees.
We are constantly improving our models. We update forecasts at midnight ET on a periodic basis.
Our models are optimized to forecast in the fields of economics, business, politics and geopolitics. They are also specifically optimized to forecast events.
VICO runs back testing regularly using a statistical measurement approach called the Brier Score, which measures the accuracy of forecasts. This is the industry standard for assessing forecast accuracy, and it is widely used in sports betting, competitive forecasting tournaments, and weather forecasting. Brier Scores are also the industry standard metric to measure the accuracy of probabilistic predictions. A Brier Score measures how closely one's forecasts pair with actual outcomes, and are computed as the mean squared error between predicted probabilities and actual outcomes. VICO applies Brier Score based back testing upon updating models or datasets.
Our models currently register an average Brier Score between 0.12 and 0.15 on our back test runs depending on specific domains, time frame, and the nature of the question. *Note: The VICO model is continuously refined, resulting in a generally decreasing Brier Score— i.e. getting better and more accurate—over time. Putting this into perspective: Brier Scores are measured on a scale of 0-1, lower being better. Anything below 0.25 indicates statistical accuracy. Below 0.2 means you are forecasting better than most experts. Prediction markets perform around 0.15. Superforecasters, the best forecasters in the world, perform around 0.1.
VICO is better than most experts in whatever field is being forecasted. To get to VICO's level of accuracy, one would need a very large prediction market (i.e. millions of dollars in bets by thousands or hundreds of thousands of users). That, or one would need to pay one or several of the few superforecasters in the world to spend hours performing the forecast research and orchestrating exhaustive analysis. Another way you can think about it is that VICO is currently more accurate than the most qualified experts, more than 90+ percent of the time—for a fraction of the cost.
VICO models are not optimized to simply track a prediction market. In fact, we would not want to, because in our view prediction markets hold several downsides including that they are capricious, often moving with high volatility around news events, are subject to manipulation, and appear to have insider trading at times. The VICO models forecast real world events using proprietary models which never contain prediction market information. VICO users can set up a forecast for anything at any time, whereas a prediction market requires setting up the betting pool and attracting bettors. VICO can also run forecasts to obtain historical data for context and correlation. VICO models can be rolling, meaning that they can assess a forecast going forward any arbitrary time period, say 90 days. Finally, VICO forecasts are fully transparent and defensible: we explain the reasoning and provide the sources behind our models' forecasts.
VICO performs measurably better than most human analysts, including subject matter experts, as is quantified by respective Brier Scores. We also regularly compare our forecasts directly with those of individual experts, such as former CIA analysts, to see how we do. Often we hear such experts say something along the lines of "I was surprised to see that VICO was so close to my own forecasts." There is a class of trained forecasters, sometimes called Superforecasters, that use specialized techniques for forecasting. VICO does not currently outperform Superforecaster accuracy as measured by Brier Score (and as far as we know, no AI model in the world does). We are approaching the level of Superforecasters and expect to supersede them within the next year. In general, compared to human analysts, VICO is much faster and more efficient at forecasting events. Where it can take hours for a single analyst to assess an event, VICO does so in seconds. VICO can also scale to assess any number of forecasts; the VICO model regularly forecasts hundreds or thousands of events at all hours of the day.
Having said all that about the accuracy of VICO, we also believe that the best forecasts are those which are useful and this is how we measure our models. A useful forecast is one that provides guidance to a decision maker, that means that it is accurate over time but also provides guidance in the interim between the forecast and the closure of the event. This is why we provide forecast data as a tickerized time series dataset. VICO is ultimately a tool to help people make decisions, whether to trade a stock, move a military unit or buy a house. Our models are optimized to help decision makers by providing both long term guidance of what will likely happen as well as immediate guidance about how those odds are shifting.
The VICO platform is accessible at vico.io. From there you can perform forecasts, obtain explanations of the forecasts, and execute scenarios and simulations. On the VICO platform, users can create and monitor their own fully customized forecasts. You can also set alerts which trigger when a forecast passes a certain threshold—say 50%—or when a certain trend occurs—such as a 10% increase in probability. Finally, enterprise users can access VICO forecasts directly through our API. VICO forecast data is also available on the Bloomberg Terminal at WSL PREDICT<GO> and WSL VICO<GO>.