If it is relevant for you, it is fundamental for us
Steady performance, Risk management, Diversification:
three relevant features for our customers, the three fundamental purposes of our method.
Alpha101+
The method using 100+ mathematical-statistical-computational proprietary scripts and an upstream diversification system to create algorithms with steady performance and risk management.
The Principles
100+ proprietary Scripts demanding
The way to have the right answers is to ask the right questions.
In Alpha101+ method, each algorithm is asked by more than 100 mathematical and statistical proprietary scripts that are elaborated by the research team: if the answers respect the fixed criteria, the algorithm “overcomes” the query and it moves to the next step.
At this proposal, Alpha101+ establishes that:
- The queries take place on a wide dataset free from bias;
- The script is elaborate and executed mainly in Matlab and Python to ensure fast and efficient processes;
- Each script imposes strict bonds so that the final algorithm is firm to such an extent that it is able to face the several situations markets present.
1 Algorithm = 1 diversified Portfolio
We are convinced that diversification must be carried out upstream and not afterwards: allocating investments in different strategies cannot be enough, this is why we have gone beyond.
Each algorithm we build behaves as a diversified portfolio because it submits the amount that has been located to an internal diversification process made by the algorithm itself.
This is possible thanks to the sub-algorithms composing the mother algorithm that:
- Operate on financial instruments presenting an almost null correlation degree between them;
- Work on different trading models;
- Have the same weight within the meta-algorithm.
The method in short
Alpha101+ consists of five stages leading to the validation of all the algorithms that are elaborated in Radiqant.
Passing a stage means to put a guarantee seal on a keystone aspect that each strategy must consider:
1 To find and confirm the presence of statistically relevant inefficiencies
2 To build an algorithm able to exploit them
3 To manage the risk and optimise performance
4 To validate the algorithm in several contexts
5 To define an exit strategy able to preserve the profits
The stages
This is the most complex stage: everything starts from here as well as everything depends on this; on our ability to make rise statistically relevant inefficiencies on which the work of the algorithm will be founded.
We use software type Bloomberg, Matlab, R, Python, C++ and Excel to select the financial instruments on which the first analysis will be founded through the relative data-set.
Most of the scripts we use help in this stage since they underline the general features of the financial instrument in question and the first cyclical inefficiencies – where given.
As follows (only) three of the most important indicators that intervene.
| STD99 | PICKUP24 | JETSEAS |
|---|---|---|
Script built in Python, characterised by a particular modified standard deviation, it analyses the volatility of the instrument in question, and a mathematical-statistical formula creates a report focusing on first eventual inefficiencies. If testing it, the indicator overcomes the limit value, then the financial instrument will be associated to a quantitative trading model type Momentum. | Statistical-computational indicator elaborated in Matlab, which analyses the frequency through which one instrument supports reversal prices movements. If its value overcomes 1.49, the financial instrument in question can be supported by a quantitative trading model type Mean Revearting. Through a report, the script underlines the potential periods in which the instrument tends to make that reversals. | This indicator elaborates reports underlining the specific dimension about the price range that characterises a financial instrument in several fixed periods. Then, through a very fast and basic optimization, it sets a Stop Loss and a Take Profit able to ensure a constant gain percentage in the course of the years and a specific risk. If the indicator overcomes the limit value, the financial instrument will be associated to a Seasonal or Statistical Arbitrage model. |
How we choose the trading model
This is the stage in which we define the trading model to adopt according to the selected financial instrument. Radiqant focuses on models type Momentum, Mean Revearting, Seasonal, Statistical Arbitrage, Correlation Strategy and Cointegration Models.
In many cases, considering the limit values, the controls and the very strict and qualitative bonds that we pose, the instrument proves to be not appropriate to none of the elaborate models and therefore it is rejected.
On the contrary, in other cases it is possible to create hybrid models, in particular when the financial instrument provides interesting answers to more scripts in the same way.
Thanks to a back-testing process based on in-sample, out of sample concepts and further simulation Montecarlo, we create a first statistical-computational trading model able to work on the financial instrument in question.
The model is rough and basic: we need it for a first and immediate check to confirm the hypothesis, that is the presence of the cyclical inefficiency.
At this point, the algorithm passes the most important test to continue the work: the indicators intervening in this stage pose strict bonds in order to ensure us that the inefficiency is capable to permit us to generate constant results in the course of time.
Here following three among the most important scripts we use.
| WR66 | ROBUST100 | DELTACOST |
|---|---|---|
Based on a complex statistical mechanism, it confirms that the profit trades are at least of 60-66%, setting up a Risk/Reward at least equal to 1:1. | It is one of the most complex indicator: it ensures us that WR66 worked and elaborated the data on a sample so robust to confirm qualitatively the win ratio higher than 60%. | It measures the difference between the commission spread costs and the performance obtained by the algorithm. Only the algorithms whose costs are not higher than 15-20% of the performance, pass the test. |
If the algorithm passes the tests in the previous stage, we proceed through further analysis about the parameters of the underlying financial instrument: the purpose is to identify those parameters able to generate the best performance in the course of time, maintaining the drawdown under the fixed limit value.
To increase the accuracy level, we make back-test with the out-of-sample data, paying attention to divide our dataset into three different parties:
Training Set: to optimise the parameters quality within the algorithm;
Test Set & Validation Set: to test the algorithm in several contexts and get confirmation of our results.
The optimisation models examine a very high number of combined variables such as Profit Factor, Take Profit, Stop Loss, Drawdown e Sharpe Ratio.
The algorithm will face two selections before concluding this stage:
| FIRST SELECTION | SECOND SELECTION |
|---|---|
In the first selection we consider only the combinations of parameters that give back a Sharpe Ratio at least higher than 2,10 and a specific minimum Average Yield per operation. | The second selection, given that the drawdown must not be overcame, will consider the best combination of the remaining parameters (in particular the Average Yield per operation and Stop Loss). |
The definitive validation of each algorithm occurs through three different levels:
| 1° Level of Validation | 2° Level of Validation | 3° Level of Validation |
|---|---|---|
|
The algorithm must produce similar results between each other in the remaining parts of the dataset and above all compared to the optimisation results. As a confirmation of this, we use Closer Parameters able to confirm the results’ closeness. |
The algorithm is submitted to a version of the method Montecarlo elaborated internally and that involves not less than 10.000 simulations. In this test, and in every single configured scenery, the drawdown must maintain under a defined probabilistic limit. |
The algorithm is submitted to a back-test for a minimum of 15 years on Validation Set, comparing performances and drawdown with the Benchmark – S&P500. The algorithm will be considered valid if every single year would have overcome the benchmark of at least 2,5% including costs, with a risk maximum equal to the benchmark one. |
We believe in the result we obtain by extremely focusing on it, but we consider unwise to make it dazzle us. According to us, the risk management also lies in understanding when an algorithm no longer performs as it should do.
This is the reason why we assign at each algorithm a complex mathematical-statistical coefficient able to let it monitor by its own in the execution, and to interrupt when the results are not in compliance with the expected ones.
Here, the coefficient assumes values from 0 to 1 and takes into consideration a huge quantity of parameters, among which there is positive monthly pay, win ratio, take profit reached, Stop Loss, monthly percentage and so on.
Each week the algorithm checks the coefficient: if it is under the value 0.5, the algorithm stops to execute orders and passes to the technical team for revision.
Alpha101+ is founded on the research, on the vertical competences and on the mathematical precision.
The model in stages and the proprietary scripts groups that are involved in each stage, ensure that our algorithms:
- Operate on inefficiencies that have strength and profitability which are verified more times in several contexts and from different points of view;
- Are built with the aim to generate a steady performance, using in the most efficient way the features of the analysed instrument;
- Are able to interrupt themselves, so that they do not damage more than expected during the building process;
Moreover, all the algorithms make an upstream diversification, since they are made of non-correlated sub-algorithms which are different in the adopted trading model and with the same relevance within the algorithm.