Nobody knows the future.
Technical analysis knows the past.
Nowcaster knows the now.
From chaos to consensus, the future is now.

What do we do?

We analyse relationships between markets. These relationships can be pairs, such as the same currency pair listed on two exchanges or two currencies with very strong long term agreement. Or they can be triangles, where AAA-BBB = (CCC-BBB / CCC-AAA).

We blend these pairs and triangles to find the consensus price.

We blend a lot of pairs. We blend a lot of triangles. Many dozens for popular markets, many thousands across all markets.

Why do we do it?

The consensus price has better latency and noise characteristics than any existing market. It is our best prediction of the current consensus price. It is the nowcast, to be traded against.

Does it work?

Yes. The Law of One Price is highly effective and consistent over the medium to long term. At tick level there are short lived, tradable price errors. These errors almost always fall away over time. In some markets the errors can last for multiple seconds and even minutes. In others they rarely last half a second – but they exist.

What can we do with it?

An extreme version of unhedged statistical latency arbitrage.

Short term error trading, scalping, market making, umbrella trading. The concept is simple: buy at a discount and sell at a premium.

How does this affect market efficiency?

By exploiting short terms price errors, you aid in price discovery and help correct market prices.

As a market maker, you are able to offer tighter spreads by using a less noise, lower latency reference price. Also you can provide extra liquidity on one side, when other makers don’t know the consensus mid price.

As a liquidity taker, you are helping to move the mid price into line with other markets.

Unlike traders who attempt to gamble on the future market price, you are being paid to do a useful job. You are the invisible hand in the market.

How does this compare to cointegration?

Cointegration techniques such as Engle and Granger 2-step and Vector Error Correction Model (VECM) can work well, but we believe they have two flaws.

Firstly they are computationally difficult, which limits the number of relationships to compare. We compare a lot of market relationships.

Secondly they don’t take into account both the entry and exit positions. We do.

We have done extensive autocorrelation analysis, cross correlation, dynamic time warping, along with other techniques. None of these give us the results that we can get from backtesting.

Do you use artificial neural networks, deep learning?

No. We have run deep learning approaches and we have not found them to succeed. Neural networks encounter the curse of dimensionality due to the large number of market inputs. Also gradient descent fails with the extreme noise of the input data (e.g. 50.2% win, 49.8% lose across many simulated trades). Stochastic Gradient descent is insufficient when the inputs are too noisy. We have solved this problem. While the implementation looks similar to a neural network, our proprietary solution can cope with extreme noise.

The approach has been complex to optimise, but the results are human interpretable.

If it is based on some kind of backtesting grid search, is this computationally feasible?

Yes. We optimised backtesting to the point where we can run vast numbers of tests, apply MLE to the output and then combine in an approach similar to additive regression.

Do we run an investment fund?

No, at least not one open to the public. We currently lack the regulatory authorisation to trade investor capital.

Do we provide an Expert Advisor, Signal or Copy Trading service?

No. We currently lack the regulatory authorisation to provide investment advice to the public.

What do we provide?

We simply provide our nowcast of the current prices, our best estimate of bid and ask. This can be used as an input to a trading bot, allowing the user to set their own trading logic and risk management.

How does this compare to other, popular indicators?

We are skeptical of applying chart analysis and technical indicators to developed markets.

Therefore, we do not believe in trading based on moving average convergence divergence (MACD), relative strength index (RSI), on-balance volume (OBV), Aroon oscillator, stochastic oscillator, Bollinger Bands, Fibonacci lines, average directional index (ADX) or average true range (ATR).

The efficient market hypothesis is quite dependable over minutes, seconds and even fractions of a second, depending on the market. This applies within the holding time of a position.

Markets already price in the available information, so medium term forecasts of future prices based on past values of the same timeseries tend to fail over the long term.

In the words of Financial Times journalist Katie Martin, our aim is to put the vomit back in the camel.

Nowcasting is different.

We do not pretend to predict the future. We aim to build the most accurate assessment of current market consensus, with no lagged inputs.

For the time that we hold a position, we have no opinion on whether the market consensus price will move up or down. We just aim to buy at a discount and sell at a premium.

Why do we focus on unhedged strategies instead of a market neutral approach?

Hedging requires being on both exchanges. It requires shorting ability, which reduces the markets that you can trade. Alternatively hedged spot trading requires holding both currencies on both exchanges at all times. This is not market neutral because you are constantly exposed to price risk.

We think of this like a used car salesman. He buys cars at below market value and then sells them above market value. Usually he does not know if the market value of that car will increase or decrease while he holds it. He accepts this market risk and profits from the spread between his buy and sell price.

To smooth out the market risk it is necessary to keep position sizes very small relative to capital. Total position size should also be small relative to capital, due to the correlated nature of financial markets during high volatility.

Is Nowcaster available to the public?

Not yet, but you can keep in touch using info@nowcaster.io