Prediction Types

Gravitan offers variety of predictions depending on your needs for many exchanges and also markets as a subscription

For each market, we offer a number of prediction types that capture different aspects of the market's operation for example price movement prediction or sideways market detection. Under each prediction type there may be more than one prediction. For example, a sideways market predcition with a future time horizon of 60 seconds or 120 seconds.

The following sections provide detailed information on prediction types and the parameters of each type that allows us to offer a range of predictions.

Price Movement

Price movement predictions forecast the direction of an asset's mid-price and minimum change in its magnitude over a pre-defined future horizon. Price movements typically forecast small changes in mid-price over short horizons of 5-30 seconds. The quality of the predictions are higher for shorter horizons and for smaller changes.

Price movement predictions are suitable for high frequency traders that take advantage of small changes in the market in short periods of time and should be considered as additional input into their trading strategy.

Market makers should also benefit from price movement predictions as additional input into their market making strategy to maintain the required level of activity and spread while minimising the risk of loss due to unfavourable market movements.

Technical Description

The direction of the asset price $(d)$ over a future horizon $(h)$ is said to be 'up' (+1) if the average relative change in the mid-price of the asset quote is forecast to exceed a predefined threshold $(\alpha)$, the direction is 'down' (-1) if the average relative change in the mid-price of the asset is forecast to be less than the negative of the pre-defined threshold $(-\alpha)$, and is forecast to be 'stationary' (0) otherwise.

$$d(h, \alpha) = \begin{cases} +1, & \frac{\Delta p}{p_0} \ge \alpha \\ \phantom{+}0, & -\alpha < \frac{\Delta p}{p_0} <\alpha \\ -1, & \frac{\Delta p}{p_0} \le -\alpha \end{cases}$$

The change in price $(\Delta p)$ is calculated as the average of the mid-price change over the horizon $h$

$$\Delta p = \frac{1}{h}\sum_ {i=1}^{h}{p_i}, ~~~i \in \N$$

The initial mid-price is denoted by $p_0$ whilst it is natural to use the mid-price of the asset at the time of prediction, we find that using a mid-price average over a short period of time $(l)$ before prediction gives more stable results.

$$p_0 = \frac{1}{l}\sum_{i = 0}^{1-l}{p_i},~~~~i\in\Z$$

Example

Consider price movement predictions for BTC with a threshold of $\alpha=0.2$ pip and a forecast horizon of $h=10$ seconds. If the initial price of BTC is $20,000, where the average price of BTC over the next 10 seconds exceeds $0.4, the price movement is reported as going up.

Sideways Market

Sideways market predictions detect a market condition where the price of an asset remains in a relatively narrow range. In a sideways market, there is limited upward or downward movement, resulting in a flat or horizontal price pattern on the price chart.

Detection of a sideways market can help you adapt your trading strategy to this type of market condition. You may want to avoid trading during this period if your strategy is to wait for a breakout or you may choose to perform range trading taking advantage of predictable the market fluctuations.

Technical Description

A market is said to be in a sideways condition $(s=1)$ when the change in the asset's mid-price over a future horizon $(h)$does not exceed a given threshold $(\alpha)$

$$s(h, \alpha) = \begin{cases} \phantom{+}0, & \exists i \in(0, h) \text{ and }|\frac{ p_i - p_0}{p_0}| \ge \alpha \\ \phantom{+}1, & \forall i \in (0, h) \text{ and } |\frac{p_i - p_0}{p_0}| <\alpha \end{cases}$$

where $p_0$ is the initial mid-price at the time of prediction and $p_i$ is the mid-price at some time $i$ within the future horizon $h$.

We offer sideways predictions for a range of thresholds (from 10 pip to 50 pip) for a range of future horizons (from 1 minute to 5 minutes). This allows you to further adapt your strategy to the level and length of a sideways market.

Example

Consider sideways predictions for Tesla shares (TSLA) with a threshold of 50 pip for the next 5 minutes. If the price of TSLA at the time of prediction is $400 and the market is predicted to go sideways, we expect the price of TSLA to fluctuate between $398 and $402 within the next 5 minutes.

Price Jumps (Coming Soon)

These predictions offer an advanced warning of an impending price jump within a predefined future horizon (typically in the order of 60 seconds). Price jumps happen infrequently. There may be no price jump predictions for several days. However, occasionally, you may receive up to a handful of predictions within a day.

We publish price jumps with and without the direction of the jump. Price jumps can be more accurately predicted without direction.

Non-directional price jump predictions are useful for market makers who typically make assumptions about continuous and smooth operation of the market and are adversely affected by large swings. They allow market makers time to close their positions temporarily until market conditions return to normal.

Price Swing (Coming Soon)

Price swing predictions provide an estimate of the upper bound and lower bound of changes in the price of an asset within a pre-defined future horizon. Price swing estimates are more informative compared to price movement predictions. However, there is usually a higher confidence in movement predictions.

Consider using a price swing prediction where the limits of market movement are more important than a simpler directional prediction.

Next Trade (Coming Soon)

Next trade predictions estimate the duration and magnitude of the next trades over a pre-defined future horizon.

Next trade predictions are particularly useful in marking making strategies where next trades and volumes may trigger trades on market makers' limit orders. Market makers would typically want to have control over the volume and timing of their trades and avoid unfavorable executions.

Volume Change (Coming Soon)

Volume based predictions cover forecast for actionable liquidity in the market over a defined time window