We provide subscriptions to live feeds of our forecasts on market data, at the moment this is only the price of $SPY, forecasting the next 10 minutes, updated every minute. We can also provide custom data feeds for forecasts of the equities you choose, business consulting services, or exclusive access to any of these data services.

Due to demand, the next live forecast will be BTC/USD. Please submit the frequency you'd like in the contact form if interested in this. It will be behind a paywall and access will be available by querying with an API key that you will get when you subscribe.
Live Update

Markov Chain Monte Carlo Forecast using Bayesian Structural Time Series Model and Hamiltonian Monte Carlo Fitting Technique

Chart is live: updates on browser refresh with new data every minute.

For tabular data that is current: Download Live Data

Data headers are: Time UTC, Price, Symbol, 1min prediction, 2 min prediction ...10 min prediction, 1 min standard deviation, 2 min standard deviation ... 10 min standard deviation.


What really matters when it comes to predicting the future is 'How do you know the model is going to be right?' and 'How long until it is unusable because it is wrong?'

"Never believe a prediction that does not empower you."

Sean Stephenson

Due to the high cost and proprietary nature of AI as a service, we do not think most commercial options empower you. They offer a black-box, one-size-fits-all option to fit your data. You may incur high computational costs in training, development, and deployment. And because of overfitting, the training and development will continue forever. The worst part is that you are being charged for the use of intellectual property that is really free, with half-baked estimates of uncertainty. Or even none at all.

At MCMC Forecast, we only utilize developers who have a background in design of experiment, hyopthesis testing, and statistics. Stick around even if you have been discouraged by predictive models in the past or are skeptical that machine learning and artificial intelligence can deliver actionable results in fintech. On the other side of the spectrum, perhaps you imagine that it's trivial to use AI to predict the future. It is not trivial. The truth is that to use open-source software that can build great models like the ones you will find here it is, in difficulty, somewhere between reading the manual for every subroutine in every software package you can imagine and copying someone's code from a tutorial. Everything you will find on here has been made with open source software and you could certainly try to figure out how to make these models yourself. We are confident that you will find it preferable to subscribe to our data. What you will find here is all well-vetted, we never crowdsource wisdom, and we are not a middleman for any products.

We provide quantified uncertainty with our forecasts and we do this by using Bayesian Structural Time Series models that we fit with Hamiltonian Monte Carlo. The Markov Chain Monte Carlo methods we use are regarded as the gold standard for Bayesian inference, plus we tried everything else and have not been more impressed by other techniques yet. Variational Inference has a place for some interesting applications that will be added at a later date, but it has increased error relative to MCMC methods. And, finally, yes we test these models live with our own money as a component of experimentaly validating them and to ensure there's a quantum of utility. Testing with actual capital provides a very limited window into the uncertainty of the model compared with how these models actually work, fundamentally, but it is quite satisfying and we see it as necessary part of understanding our users' needs. On this website, we do not provide or recommend strategies except for some occassional blog posts to stimulate thought. This is not a robo-advisor or stock trading robot, it's just a reasonably priced data feed based on state-of-the-art, open-source software.

We are confident you will like what you find here. And we will be updating with blog posts on measurement uncertainty, propagation of errors, and interesting data-processing techniques. For now, have a look around and enjoy playing with our backtest data from the running model. There are no implied warranties of uptime, accuracy, or profitability from the publicly available data feeds. If you find a good strategy based on the running model, you're welcome to use it at your own risk and can contact the site for any other information you'd like about subscribing to a service with a better guarantee of uptime.

Below is a comparison of the Bayesian Structural Time Series model's forecast errors at each 10-minute prediction scaled to units of 1 predicted standard deviation and the error from subtracting a linear baseline from the actual prices (the standard deviation was ~$22).

I would prefer to work with the data on the right.
Histograms of the MCMC technique and a linear baseline, i.e. stocks go up approach, with MCMC forecasting we can make the distribution normal

Download the historical data & predictions (you can recalculate anything shown on this page from this data):Download


If you would like to subscribe, please wait until forecasts behind a paywall are added. If you subscribe early it will be considered a donation to support costs of development and hosting.

Current Price: $5/month
Renews until canceled and includes 1-time account setup fee of $25

FYI! Use the contact form below to initiate all other requests for services or consulting. Options include custom financial or risk analysis, currency pairs, inventory management, etc.

Contact Let us know what you think, if you find a bug, or would like to initiate a project: