Please, Contact Us if you have any inquiry about Exchange Integration.
Our Risk Management strategies, together with our Real-Time Performance Monitoring technologies are responsible for not letting any potential losses scale up – On speculative trading algorithms, this may happen really fast. A rock solid cooperation between these subsystems is key for our sane 100% automated Algo Trading operation.
When using our Goal Oriented Risk Manager, the Ogre Robot platform is able to predict on which scenarios a given Trading Algorithm is able to achieve that goal or not, as well as informing the probability for each scenario happening, one after another. The pre-rendered scenario sequences may cover the whole day or just the next minutes, depending on the market behavior. Human input may restrict further were to stop issueing orders, based on the predictability window and the maximum hold time an asset may have in order to fulfill the goal.
Many automated models trade nowadays. Speed helped us here: about predictability, our fast platform allows us to migrate a little further from the realm of probabilities to the realm of certainty. Ogre Robot has a built in Automatic Risk Manager which uses Computational Intelligence Technologies and is able to come up with on-the-fly strategies, both on simulation and operation times.
Identifying interesting events is as simple as setting triggers on conditions. E.g.:
For instance, we may wakeup routine SELL1 when the ‘ask’ of security 1 is rising for the last 5 successful transactions, it is greater than security 2’s ‘ask’ and the ‘bid’ for security 3 is dropping for, at least, 5 consecutive transactions.
Usually, after a wake up the algorithm is able to use more complex operations in order to check if the pretended action (SELL1 in the previous example) should really sell security 1 – like determining a possible bid for it. This involves a process of decision making and it may be done without fear: in case of a certain number of adverse operations (yielding revenue loss), the Risk Manager in charge will stop the operation. The routine may, then, be reprogrammed with the ease of log files, scenario dumps, etc. The same rules apply whether in simulation or real operation mode.
Implemented Models may be as elaborated as C, C++, Java or Python languages allow. For those who need customized training, we have a bunch of APIs to feed the algorithm with real old data – Order Book, Trade Book, … If present, the algorithm’s custom training method will coordinate the process and may require any number of replays (simulating real-time, but actually taking place faster than real-time), may set signal matrices (useful for neural networks) and command new training processes to start in parallel with different settings, in the search for the optimum values – or simply use our built-in Genetic Algorithm.
Ogre Robot platform is very modular. This means the built in monitoring facility may be extended to watch on whatever data your algorithm produces or rely on and issue warnings, trigger the safe mode and use any of the API methods. The monitoring may be running on the same machine as the operation or be set remotely, analyzing independent data, for an extra level of protection.
These algorithms explore patterns, cycles, corelations and other predictable behavior infered from observing market raw data. They may implement human knowledge or use computational intelligence to find such patterns. There are many possible patterns and many possible ways of detecting them, therefore there are several algorithms, ranging from high to low recurrence.
These algorithms aim to make gains at the milli or even micro second range, hence the term High Frequency. They are the messy warriors big financial institutions build to gladiate one another, providing liquidity and consistent related securities prices as a side effect. They need to be fast, so they are not very smart. Usually they follow very simple rules and are co-located at the same data center as the stock exchange, since every µs counts on their race.
Market behaviour that may be represented by mathematical or statistical models are often borned in theory and later validated in practice. They form the model-based class of algorithms. Many of them are possibilities that either cannot be observed or are valid only on specific market conditions.