Hunting unique investment opportunities with artificial intelligence and taking into account different risk measurements makes DFO crypto funds a powerful fully automated hedge funds.
We provide two self-driven quantitative trading machines based on Spot and Futures markets that uses Deep Learning and Machine Learning models plus more than 500 features for signal generation, risk management, and order execution.
Here, we provide the performance overview and statistical comparison between DFO Funds, Barclays Crypto index and Bitcoin.
Reported performance in these figures consists of back-test results as well as system forward test. This is due to the fact that several metrics, i.e. Sharpe Ratio, are only meaningful with a certain amount of historical data.
All results period: March 31, 2020, to March 31, 2023.
Reported results before July 2021 and November 2022 are backtest for DFO Spot and DFO Futures Funds, respectively. After that the results are generated with our full featured platform.
Artificial intelligence takes investment decisions here. Rule-based investing – fast, quantitative, reliable, and inexpensive
DFO as an algorithmic data provider uses artificial intelligence and data analysis to create strategies for trading in the cryptocurrency market.
These strategies are constantly optimized and aim to obtain high returns compared to risk.
The DFO crypto pipeline is implemented in four steps. Each of these steps uses its own unique technology.
All DFO models are active in the spot market and all steps DFO order execution models automatically in the clients’ accounts.
The most important part of any model is order generation which is including the coin selection and entry timing. This part of DFO crypto consists of two phases:
First, 65 coins are given to the coin selection model in terms of the largest market size. Every 12 hours, the model selects suitable candidate coins for trading in the next 12 hours based on their recent movement/trend and volatility.
Open a position:
To generate an order on the selected coins, the Al model checks the past state of the coin and the market situation. If the coin momentum triggers the threshold of the risk to return, the order is sent to the order opening section.
Every 1 hour, the model checks more than 500 parameters on the selected coins and decides whether to enter the market or not. According to the state of the coin, the appropriate entry point to open the position is determined.
Since our strategies are completely systematic, we have developed most of our investment technologies in-house. This makes it possible to control the quality of execution and meet the requirements of our trading algorithms. These internal automated tools help our quantitative research team quickly and efficiently identify new sources of trend in large data sets.
The first step in deploying a new quantitative investment strategy is data acquisition. Today, advances in technology and declining storage costs have dramatically increased the amount of data stored by organizations across all industries. To support our Al model to find a good price level to generate orders, we use big data, browsing multiple datasets from different areas. They include data such as historical price data, indicators, price action patterns and more complex alternative datasets. We deliver 500 factors to the Al model as input.
Since we are using different sources of data, such as order books, news, indicators, and technical and fundamental data, we may end up with unstructured format that is not suitable for the Al models, so data needs to be validated and cleaned. Our Data team has implemented various automated tools to support big data that need to be cleaned up and checked for potential errors. These instruments help our models to analyze and reduce the appearance of erroneous data that can lead to biased testing on historical data and erroneous conclusions.
Once the data is cleaned, our Al model as called Carbon begins to explore the data. Carbon has six sub-models including two deep neural networks and four classic machine learning algorithms. Carbon creates an ensemble learning framework in order to improve the accuracy and enhance the return. As a result, Carbon generates Buy, Sell, and Hold signals on each coin.
After discovering new sources of alpha signal, our researchers conduct backtests that span long historical data to increase the likelihood of finding reliable signals. The testing stage takes a lot of time due to the large amount of historical data on which the strategy can be run and it needs an intelligent design to avoid overfitting. Our quantitative backtest model requires advanced algorithms and sufficient processing power to complete their tests as quickly as possible.
One of the most important factors in making a profit from a strategy is choosing the exit point wisely. Exiting early or late can wipe out all the potential profit of a position.
Hard SL & TP
According to the history of the coin, the level of volatility of each coin and the maximum amount of acceptable risk, stop loss, and take profit are determined. (The probability of overfitting in determining these parameters is high, and the optimization process needs further attention)
Build Risk-Free Position
If the stop loss is activated, the entire position will be closed, while if the take profit level is touched, 50% of the position will be closed and the stop loss will be moved to the entry point (breakeven).
Trailing Stop Loss
In DFO, we have an artificial intelligence model that trails stop loss based on the trend of the coin and the market. In main trends, the trailing stop moves with the market and prevents early exit from the position. Finally, when this stop loss is activated, the remaining 50% of the position will be sold and the position will be closed.
The size of the order comes from a model that allocates a predesigned risk function based on the investors’ preferences.
At any given time, if multiple orders are generated simultaneously, the maximum number of positions that can be opened is based on the past performance of the model on that coin and the confidence level of the Al model for that order.
If the number of orders exceeds the maximum selection limit, the selection between coins is made based on the model historical performance on that coin. The share of each order from the available base asset is determined based on the coin’s volatility in the market and the coin’s trading volume in the last 12 hours.
DFO has built the execution system according to its own needs. This system connects to the exchange through API and places orders.
Currently, it works on Binance and Kocoin, but due to its infrastructure, it can easily be developed and connected to other exchanges. This system has the possibility of executing all the commands generated in the previous section, such as risk-free position, trailing, etc.
Every investment is associated with risk. As Deep Finance team, we aim to increase your chance of winning against the market by using Al based technologies and minimize your risk by our intelligent risk management method. Our provided Sharpe ratio and Sortino ratio indicate our potential and show outstanding risk adjusted performance in the market.