Overcome and Analysis

While Overcome provides the operation that fits for every OHLCV row, in a history input data set; Analysis provides the profits and number of overlapped opened positions for a given data set, with fulfilled operations.

Given a constant take profit and a stop loss, and a time base sorted data vector with the "high", "low", and "close" values from a stock market product, Overcome calculates the potential earnings in both, buying and selling for every step in the timeline.

Analysis by the other hand is a calculation after an operation has been already set. It counts the overlapped buying and selling positions, separately, and the profits for both as well.

Install

Install from Pypi repository from at least a Python 3.7.

pip install overcome

Overcome usage

Instantiate Overcome with the constant value for take profit and stop loss. A value for the precision threshold is required as well. The precision threshold applies on whether a "close" value is close to the take profit or stop loss boundaries.

outcome = Overcome(
        threshold=np.float32(0.00001),
        take_profit=np.float32(0.001),
        stop_loss=np.float32(0.001)
    )

Call apply method with the input and wait for the output. The required input structure is a numpy array with a shape of (LENGTH, 3).

The order of the 3 columns is very strict. The first one is for the "high" values, the second one for the "low" values and the third one for "close" values.

earn_buying, earn_selling = outcome.apply(high_low_close)

It returns a tuple of two arrays with the same number of items than the input, indeed matching the same timeline. The first array is for the earnings from buying positions and the second one for the selling ones.

Opened positions limit

Opening all positions without limiting their number may be pretty risky. Overcome Simulation supports to set a predefined positions number limit in each direction, buying and selling. So, for example, if the positions limit is set to 10, when the first 10 positions will be opened in the simulated outcome, the following ones won't open, and then they are not accountable for profit or loss.

Creating an Overcome instance with opened positions limited at 10 for buying and 10 for selling is as follows.

outcome = Overcome(
        threshold=np.float32(0.00001),
        take_profit=np.float32(0.001),
        stop_loss=np.float32(0.001),
        positions_limit=10
    )

Counters and maximum delay

Boolean has_counters is an optional constructor argument to ask for a counter for each, buying and selling positions, with the number of steps between the position starting and closing steps. These optional information comes in the returning tuple as two new items.

outcome = Overcome(
        threshold=np.float32(0.00001),
        take_profit=np.float32(0.001),
        stop_loss=np.float32(0.001),
        positions_limit=10,
        has_counters=True
    )
earn_buying, earn_selling, buying_lengths, selling_lengths = \
    outcome.apply(high_low_close)

Having the number of steps of every operation lets us filter out the longer operations in order to get only the fastest ones, apparently the most accurate ones. The integer type optional argument max_delay makes to ignore the earnings from the operations longer than the maximum steps set in this value. In the following example, the earnings from the positions with more than 10 steps will be ignored and set to 0.

Setting a value over 0 to max_delay forces has_counters to enable in order to get the required counts. So, the returning tuple length is 4 as well.

outcome = Overcome(
        threshold=np.float32(0.00001),
        take_profit=np.float32(0.001),
        stop_loss=np.float32(0.001),
        positions_limit=10,
        max_delay=10
    )
earn_buying, earn_selling, buying_lengths, selling_lengths = \
    outcome.apply(high_low_close)

Input from a dataframe

Starting with a Dataframe as df from any product historical data, convert the columns into the required input by the following expression.

high_low_close = df[["high", "low", "close"]].to_numpy(dtype=np.float32)

Then apply the calculation and merge the result into the original Dataframe as follows.

(df["earn_buying"], df["earn_selling"]) = outcome.apply(high_low_close)

Example

The following table is an example of dataframe with the new columns for buying and selling earnings already calculated. This table is part of a test suite, so the data is not real, it is just prepared to achieve a scenario with a variety of positive and negative earnings in both operations, buying and selling.

The configuration for take profit for this example is 0.001 and the stop loss is 0.0007.

timestamp	close	high	low	earn_buying	earn_selling
2022-04-01 00:01:00	1.0	1.0	1.0	0.001	-0.0007
2022-04-01 00:02:00	1.0	1.0	1.0	0.001	-0.0007
2022-04-01 00:03:00	1.0005	1.0007	1.0	0.001	-0.0007
2022-04-01 00:04:00	1.0005	1.001	1.0001	0.001	-0.0007
2022-04-01 00:05:00	1.0006	1.001	1.0002	-0.0007	-0.0007
2022-04-01 00:06:00	1.0007	1.0011	1.0	-0.0007	-0.0007
2022-04-01 00:07:00	1.0008	1.0012	1.0004	-0.0007	-0.0007
2022-04-01 00:08:00	1.0009	1.0013	1.0005	-0.0007	0.001
2022-04-01 00:09:00	1.001	1.0015	1.0006	-0.0007	0.001
2022-04-01 00:10:00	1.001	1.0015	0.9995	-0.0007	0.001
2022-04-01 00:11:00	1.001	1.0015	0.9993	-0.0007	0.001
2022-04-01 00:12:00	1.001	1.0016	0.999	0.0	0.0

Analysis usage

Instantiate Analysis by giving an integer type constant for every distinct operation: relax, selling, buying.

analysis = Analysis(
        threshold=np.float32(0.00001),
        take_profit=np.float32(0.001),
        stop_loss=np.float32(0.001),
        categories={
            "relax": 0,
            "sell": 1,
            "buy": 2
        }
)
results = analysis.apply(predictions, ohlcv_dataframe)

Minimizing costs

There is a service that in a regular alternate way, sets off the given active positions like buying or selling. This service is an optional dependency of Analysis, so it allows to analyze the given operations, by minimizing them in the way on.

The following example, sets off 2 of 3 positions. Starting at the second row of the given operations due to the RegularCostMinimizer's offset optional argument.

RELAX_CATEGORY = 0

analysis = Analysis(
        threshold=np.float32(0.00001),
        take_profit=np.float32(0.001),
        stop_loss=np.float32(0.001),
        categories={
            "relax": RELAX_CATEGORY,
            "sell": 1,
            "buy": 2
        },
        minimizer=RegularCostMinimizer(RELAX_CATEGORY, 2, offset=1)
)
results = analysis.apply(predictions, ohlcv_dataframe)

Test and development

This project is based on BDD development. Any change starts here tests/features.

Moreover, you can find the most specific documentation about any module and package of this project.

Tests require behave and pandas. Execute the command below from the source code root directory to install the extra requirements to start developing on this project.

pip install -e ".[dev]"

PYTHONPATH must contain the main source directory ./src to execute the tests. Use the fantastic command as follows.

PYTHONPATH=./src behave tests/features

There are a few integration tests as well. Execute as follows.

PYTHONPATH=./src python tests/integration/test_overcome.py

Author

Jaume Mila Bea [email protected]