Пример #1
0
    def simulate_future_prices(self, observation_date, requirements, path_count, calibration_params):
        # Compute correlated Brownian motions for each market.

        if not requirements:
            return

        all_brownian_motions = self.get_brownian_motions(observation_date, requirements, path_count,
                                                         calibration_params)

        delivery_dates = defaultdict(set)
        for requirement in requirements:
            fixing_date = requirement[1]
            delivery_date = requirement[2]
            delivery_dates[fixing_date].add(delivery_date)

        # delivery_dates[observation_date].add(observation_date)

        # Compute simulated market prices using the correlated Brownian
        # motions, the actual historical volatility, and the last price.
        for commodity_name, brownian_motions in all_brownian_motions:
            # Get the 'last price' for this commodity.

            index = calibration_params['market'].index(commodity_name)
            sigma = calibration_params['sigma'][index]
            curve = ForwardCurve(commodity_name, calibration_params['curve'][commodity_name])
            for fixing_date, brownian_rv in brownian_motions:
                for delivery_date in sorted(delivery_dates[fixing_date]):
                    forward_price = curve.get_price(delivery_date)
                    T = get_duration_years(observation_date, fixing_date)
                    simulated_value = forward_price * scipy.exp(sigma * brownian_rv - 0.5 * sigma * sigma * T)
                    yield commodity_name, fixing_date, delivery_date, simulated_value
Пример #2
0
    def simulateFuturePrices(self, market_names, fixing_dates,
                             observation_time, path_count, market_calibration):
        allBrownianMotions = self.getBrownianMotions(market_names,
                                                     fixing_dates,
                                                     observation_time,
                                                     path_count,
                                                     market_calibration)
        # Compute market prices, so the Market object doesn't do this.
        import scipy
        all_market_prices = {}
        for (marketName, brownianMotions) in allBrownianMotions.items():
            lastPrice = market_calibration['%s-LAST-PRICE' %
                                           marketName.upper()]
            actualHistoricalVolatility = market_calibration[
                '%s-ACTUAL-HISTORICAL-VOLATILITY' % marketName.upper()]
            marketPrices = {}
            for (fixingDate, brownianRv) in brownianMotions.items():
                sigma = actualHistoricalVolatility / 100.0
                T = get_duration_years(observation_time, fixingDate)
                marketRv = lastPrice * scipy.exp(sigma * brownianRv -
                                                 0.5 * sigma * sigma * T)
                marketPrices[fixingDate] = marketRv

            all_market_prices[marketName] = marketPrices
        return all_market_prices
Пример #3
0
    def simulate_future_prices(self, observation_date, requirements,
                               path_count, calibration_params):
        # Compute correlated Brownian motions for each market.

        if not requirements:
            return

        all_brownian_motions = self.get_brownian_motions(
            observation_date, requirements, path_count, calibration_params)

        delivery_dates = defaultdict(set)
        for requirement in requirements:
            fixing_date = requirement[1]
            delivery_date = requirement[2]
            delivery_dates[fixing_date].add(delivery_date)

        # delivery_dates[observation_date].add(observation_date)

        # Compute simulated market prices using the correlated Brownian
        # motions, the actual historical volatility, and the last price.
        for commodity_name, brownian_motions in all_brownian_motions:
            # Get the 'last price' for this commodity.

            index = calibration_params['market'].index(commodity_name)
            sigma = calibration_params['sigma'][index]
            curve = ForwardCurve(commodity_name,
                                 calibration_params['curve'][commodity_name])
            for fixing_date, brownian_rv in brownian_motions:
                for delivery_date in sorted(delivery_dates[fixing_date]):
                    forward_price = curve.get_price(delivery_date)
                    T = get_duration_years(observation_date, fixing_date)
                    simulated_value = forward_price * scipy.exp(
                        sigma * brownian_rv - 0.5 * sigma * sigma * T)
                    yield commodity_name, fixing_date, delivery_date, simulated_value
Пример #4
0
    def simulate_future_prices(self, market_names, fixing_dates, observation_date, path_count, calibration_params):
        # Compute correlated Brownian motions for each market and fixing date.
        if market_names:
            all_brownian_motions = self.get_brownian_motions(market_names, fixing_dates, observation_date, path_count,
                                                             calibration_params)

            # Compute simulated market prices using the correlated Brownian
            # motions, the actual historical volatility, and the last price.
            import scipy
            for market_name, brownian_motions in all_brownian_motions:
                last_price = calibration_params['%s-LAST-PRICE' % market_name.upper()]
                actual_historical_volatility = calibration_params['%s-ACTUAL-HISTORICAL-VOLATILITY' % market_name.upper()]
                sigma = actual_historical_volatility / 100.0
                for fixing_date, brownian_rv in brownian_motions:
                    T = get_duration_years(observation_date, fixing_date)
                    simulated_value = last_price * scipy.exp(sigma * brownian_rv - 0.5 * sigma * sigma * T)
                    yield market_name, fixing_date, simulated_value
Пример #5
0
    def simulateFuturePrices(self, market_names, fixing_dates, observation_time, path_count, market_calibration):
        allBrownianMotions = self.getBrownianMotions(market_names, fixing_dates, observation_time, path_count, market_calibration)
        # Compute market prices, so the Market object doesn't do this.
        import scipy
        all_market_prices = {}
        for (marketName, brownianMotions) in allBrownianMotions.items():
            lastPrice = market_calibration['%s-LAST-PRICE' % marketName.upper()]
            actualHistoricalVolatility = market_calibration['%s-ACTUAL-HISTORICAL-VOLATILITY' % marketName.upper()]
            marketPrices = {}
            for (fixingDate, brownianRv) in brownianMotions.items():
                sigma = actualHistoricalVolatility / 100.0
                T = get_duration_years(observation_time, fixingDate)
                marketRv = lastPrice * scipy.exp(sigma * brownianRv - 0.5 * sigma * sigma * T)
                marketPrices[fixingDate] = marketRv

            all_market_prices[marketName] = marketPrices
        return all_market_prices
Пример #6
0
    def simulate_future_prices(self, observation_date, requirements, path_count, calibration_params):
        # Compute correlated Brownian motions for each market.

        if not requirements:
            return

        all_brownian_motions = self.get_brownian_motions(observation_date, requirements, path_count, calibration_params)

        # Compute simulated market prices using the correlated Brownian
        # motions, the actual historical volatility, and the last price.
        for commodity_name, brownian_motions in all_brownian_motions:
            # Get the 'last price' for this commodity.
            param_name = '%s-LAST-PRICE' % commodity_name
            last_price = self.get_calibration_param(param_name, calibration_params)

            # Get the 'actual historical volatility' for this commodity.
            param_name = '%s-ACTUAL-HISTORICAL-VOLATILITY' % commodity_name
            actual_historical_volatility = self.get_calibration_param(param_name, calibration_params)

            sigma = actual_historical_volatility / 100.0
            for fixing_date, brownian_rv in brownian_motions:
                T = get_duration_years(observation_date, fixing_date)
                simulated_value = last_price * scipy.exp(sigma * brownian_rv - 0.5 * sigma * sigma * T)
                yield commodity_name, fixing_date, fixing_date, simulated_value
Пример #7
0
    def get_brownian_motions(self, market_names, fixing_dates, observation_date, path_count, calibration_params):
        assert isinstance(market_names, list), market_names
        assert isinstance(fixing_dates, list), fixing_dates
        assert isinstance(observation_date, datetime.date), observation_date
        assert isinstance(path_count, int), path_count

        # Get an ordered list of all the dates.
        fixing_dates = set(fixing_dates)
        fixing_dates.add(observation_date)
        all_dates = sorted(fixing_dates)

        len_market_names = len(market_names)
        len_all_dates = len(all_dates)

        if len_market_names == 0:
            return []

        # Diffuse random variables through each date for each market (uncorrelated increments).
        import numpy
        import scipy.linalg
        from numpy.linalg import LinAlgError
        brownian_motions = scipy.zeros((len_market_names, len_all_dates, path_count))
        for i in range(len_market_names):
            _start_date = all_dates[0]
            start_rv = brownian_motions[i][0]
            for j in range(len_all_dates - 1):
                fixing_date = all_dates[j + 1]
                draws = numpy.random.standard_normal(path_count)
                T = get_duration_years(_start_date, fixing_date)
                if T < 0:
                    raise DslError("Can't really square root negative time durations: %s. Contract starts before observation time?" % T)
                end_rv = start_rv + scipy.sqrt(T) * draws
                try:
                    brownian_motions[i][j + 1] = end_rv
                except ValueError as e:
                    raise ValueError("Can't set end_rv in brownian_motions: %s" % e)
                _start_date = fixing_date
                start_rv = end_rv

        if len_market_names > 1:
            correlation_matrix = numpy.zeros((len_market_names, len_market_names))
            for i in range(len_market_names):
                for j in range(len_market_names):

                    # Get the correlation between market i and market j...
                    name_i = market_names[i]
                    name_j = market_names[j]
                    if name_i == name_j:
                        # - they are identical
                        correlation = 1
                    else:
                        # - correlation is expected to be in the "calibration" data
                        correlation = self.get_correlation_from_calibration(calibration_params, name_i, name_j)

                    # ...and put the correlation in the correlation matrix.
                    correlation_matrix[i][j] = correlation

            # Compute lower triangular matrix, using Cholesky decomposition.
            try:
                U = scipy.linalg.cholesky(correlation_matrix)
            except LinAlgError as e:
                raise DslError("Cholesky decomposition failed with correlation matrix: %s: %s" % (correlation_matrix, e))

            # Construct correlated increments from uncorrelated increments
            # and lower triangular matrix for the correlation matrix.
            try:
                # Put markets on the last axis, so the broadcasting works, before computing
                # the dot product with the lower triangular matrix of the correlation matrix.
                brownian_motions_correlated = brownian_motions.T.dot(U)
            except Exception as e:
                msg = ("Couldn't multiply uncorrelated Brownian increments with decomposed correlation matrix: "
                       "%s, %s: %s" % (brownian_motions, U, e))
                raise DslError(msg)

            # Put markets back on the first dimension.
            brownian_motions_correlated = brownian_motions_correlated.transpose()
            brownian_motions = brownian_motions_correlated

        # Put random variables into a nested Python dict, keyed by market name and fixing date.
        all_brownian_motions = []
        for i, market_name in enumerate(market_names):
            market_rvs = []
            for j, fixing_date in enumerate(all_dates):
                rv = brownian_motions[i][j]
                market_rvs.append((fixing_date, rv))
            all_brownian_motions.append((market_name, market_rvs))

        return all_brownian_motions
Пример #8
0
    def getBrownianMotions(self, market_names, fixing_dates, observation_time,
                           path_count, market_calibration):
        assert isinstance(observation_time,
                          datetime.datetime), type(observation_time)
        assert isinstance(path_count, int), type(path_count)

        market_names = list(market_names)
        allDates = [observation_time] + sorted(fixing_dates)

        lenMarketNames = len(market_names)
        lenAllDates = len(allDates)

        # Diffuse random variables through each date for each market (uncorrelated increments).
        import numpy
        import scipy.linalg
        from numpy.linalg import LinAlgError
        brownianMotions = scipy.zeros(
            (lenMarketNames, lenAllDates, path_count))
        for i in range(lenMarketNames):
            _start_date = allDates[0]
            startRv = brownianMotions[i][0]
            for j in range(lenAllDates - 1):
                fixingDate = allDates[j + 1]
                draws = numpy.random.standard_normal(path_count)
                T = get_duration_years(_start_date, fixingDate)
                if T < 0:
                    raise DslError(
                        "Can't really square root negative time durations: %s. Contract starts before observation time?"
                        % T)
                endRv = startRv + scipy.sqrt(T) * draws
                try:
                    brownianMotions[i][j + 1] = endRv
                except ValueError as e:
                    raise ValueError("Can't set endRv in brownianMotions: %s" %
                                     e)
                _start_date = fixingDate
                startRv = endRv

        # Read the market calibration data.
        correlations = {}
        for marketNamePairs in itertools.combinations(market_names, 2):
            marketNamePairs = tuple(sorted(marketNamePairs))
            calibrationName = "%s-%s-CORRELATION" % marketNamePairs
            try:
                correlation = market_calibration[calibrationName]
            except KeyError as e:
                msg = "Can't find correlation between '%s' and '%s': '%s' not defined in market calibration: %s" % (
                    marketNamePairs[0], marketNamePairs[1],
                    market_calibration.keys(), e)
                raise DslError(msg)
            else:
                correlations[marketNamePairs] = correlation

        correlationMatrix = numpy.zeros((lenMarketNames, lenMarketNames))
        for i in range(lenMarketNames):
            for j in range(lenMarketNames):
                if market_names[i] == market_names[j]:
                    correlation = 1
                else:
                    key = tuple(sorted([market_names[i], market_names[j]]))
                    correlation = correlations[key]
                correlationMatrix[i][j] = correlation

        try:
            U = scipy.linalg.cholesky(correlationMatrix)
        except LinAlgError as e:
            raise DslError(
                "Couldn't do Cholesky decomposition with correlation matrix: %s: %s"
                % (correlationMatrix, e))

        # Correlated increments from uncorrelated increments.
        #brownianMotions = brownianMotions.transpose() # Put markets on the last dimension, so the broadcasting works.
        try:
            brownianMotionsCorrelated = brownianMotions.T.dot(U)
        except Exception as e:
            msg = "Couldn't multiply uncorrelated Brownian increments with decomposed correlation matrix: %s, %s: %s" % (
                brownianMotions, U, e)
            raise DslError(msg)
        brownianMotionsCorrelated = brownianMotionsCorrelated.transpose(
        )  # Put markets back on the first dimension.
        brownianMotionsDict = {}
        for i, marketName in enumerate(market_names):
            marketRvs = {}
            for j, fixingDate in enumerate(allDates):
                rv = brownianMotionsCorrelated[i][j]
                marketRvs[fixingDate] = rv
            brownianMotionsDict[marketName] = marketRvs

        return brownianMotionsDict
Пример #9
0
    def get_brownian_motions(self, observation_date, requirements, path_count,
                             calibration_params):
        assert isinstance(observation_date,
                          datetime.datetime), observation_date
        assert isinstance(requirements, list), requirements
        assert isinstance(path_count, int), path_count

        commodity_names, fixing_dates = self.get_commodity_names_and_fixing_dates(
            observation_date, requirements)

        len_commodity_names = len(commodity_names)

        len_fixing_dates = len(fixing_dates)

        # Check the observation date equals the first fixing date.
        assert observation_date == fixing_dates[0], "Observation date {} not equal to first fixing date: {}" \
                                                    "".format(observation_date, fixing_dates[0])

        # Diffuse random variables through each date for each market (uncorrelated increments).
        brownian_motions = scipy.zeros(
            (len_commodity_names, len_fixing_dates, path_count))
        all_brownian_motions = []

        if len_fixing_dates and len_commodity_names:
            for i in range(len_commodity_names):
                _start_date = fixing_dates[0]
                start_rv = brownian_motions[i][0]
                for j in range(len_fixing_dates - 1):
                    fixing_date = fixing_dates[j + 1]
                    draws = scipy.random.standard_normal(path_count)
                    T = get_duration_years(_start_date, fixing_date)
                    if T < 0:
                        raise DslError(
                            "Can't really square root negative time durations: %s. Contract starts before "
                            "observation time?" % T)
                    end_rv = start_rv + scipy.sqrt(T) * draws
                    try:
                        brownian_motions[i][j + 1] = end_rv
                    except ValueError as e:
                        raise ValueError(
                            "Can't set end_rv in brownian_motions: %s" % e)
                    _start_date = fixing_date
                    start_rv = end_rv

        if len_commodity_names > 1:
            correlation_matrix = scipy.zeros(
                (len_commodity_names, len_commodity_names))
            for i in range(len_commodity_names):
                for j in range(len_commodity_names):

                    # Get the correlation between market i and market j...
                    name_i = commodity_names[i]
                    name_j = commodity_names[j]
                    if name_i == name_j:
                        # - they are identical
                        correlation = 1
                    else:
                        # - correlation is expected to be in the "calibration" data
                        correlation = self.get_correlation_from_calibration(
                            calibration_params, name_i, name_j)

                    # ...and put the correlation in the correlation matrix.
                    correlation_matrix[i][j] = correlation

            # Compute lower triangular matrix, using Cholesky decomposition.
            try:
                U = scipy.linalg.cholesky(correlation_matrix)
            except LinAlgError as e:
                msg = "Cholesky decomposition failed with correlation matrix: %s: %s" % (
                    correlation_matrix, e)
                raise DslError(msg)

            # Construct correlated increments from uncorrelated increments
            # and lower triangular matrix for the correlation matrix.
            try:
                # Put markets on the last axis, so the broadcasting works, before computing
                # the dot product with the lower triangular matrix of the correlation matrix.
                brownian_motions_correlated = brownian_motions.T.dot(U)
            except Exception as e:
                msg = (
                    "Couldn't multiply uncorrelated Brownian increments with decomposed correlation matrix: "
                    "%s, %s: %s" % (brownian_motions, U, e))
                raise DslError(msg)

            # Put markets back on the first dimension.
            brownian_motions_correlated = brownian_motions_correlated.transpose(
            )
            brownian_motions = brownian_motions_correlated

        # Put random variables into a nested Python dict, keyed by market commodity_name and fixing date.
        for i, commodity_name in enumerate(commodity_names):
            market_rvs = []
            for j, fixing_date in enumerate(fixing_dates):
                rv = brownian_motions[i][j]
                market_rvs.append((fixing_date, rv))
            all_brownian_motions.append((commodity_name, market_rvs))

        return all_brownian_motions
Пример #10
0
    def getBrownianMotions(self, market_names, fixing_dates, observation_time, path_count, market_calibration):
        assert isinstance(observation_time, datetime.datetime), type(observation_time)
        assert isinstance(path_count, int), type(path_count)

        market_names = list(market_names)
        allDates = [observation_time] + sorted(fixing_dates)

        lenMarketNames = len(market_names)
        lenAllDates = len(allDates)

        # Diffuse random variables through each date for each market (uncorrelated increments).
        import numpy
        import scipy.linalg
        from numpy.linalg import LinAlgError
        brownianMotions = scipy.zeros((lenMarketNames, lenAllDates, path_count))
        for i in range(lenMarketNames):
            _start_date = allDates[0]
            startRv = brownianMotions[i][0]
            for j in range(lenAllDates - 1):
                fixingDate = allDates[j + 1]
                draws = numpy.random.standard_normal(path_count)
                T = get_duration_years(_start_date, fixingDate)
                if T < 0:
                    raise DslError("Can't really square root negative time durations: %s. Contract starts before observation time?" % T)
                endRv = startRv + scipy.sqrt(T) * draws
                try:
                    brownianMotions[i][j + 1] = endRv
                except ValueError as e:
                    raise ValueError("Can't set endRv in brownianMotions: %s" % e)
                _start_date = fixingDate
                startRv = endRv

        # Read the market calibration data.
        correlations = {}
        for marketNamePairs in itertools.combinations(market_names, 2):
            marketNamePairs = tuple(sorted(marketNamePairs))
            calibrationName = "%s-%s-CORRELATION" % marketNamePairs
            try:
                correlation = market_calibration[calibrationName]
            except KeyError as e:
                msg = "Can't find correlation between '%s' and '%s': '%s' not defined in market calibration: %s" % (
                    marketNamePairs[0],
                    marketNamePairs[1],
                    market_calibration.keys(),
                    e
                )
                raise DslError(msg)
            else:
                correlations[marketNamePairs] = correlation

        correlationMatrix = numpy.zeros((lenMarketNames, lenMarketNames))
        for i in range(lenMarketNames):
            for j in range(lenMarketNames):
                if market_names[i] == market_names[j]:
                    correlation = 1
                else:
                    key = tuple(sorted([market_names[i], market_names[j]]))
                    correlation = correlations[key]
                correlationMatrix[i][j] = correlation

        try:
            U = scipy.linalg.cholesky(correlationMatrix)
        except LinAlgError as e:
            raise DslError("Couldn't do Cholesky decomposition with correlation matrix: %s: %s" % (correlationMatrix, e))

        # Correlated increments from uncorrelated increments.
        #brownianMotions = brownianMotions.transpose() # Put markets on the last dimension, so the broadcasting works.
        try:
            brownianMotionsCorrelated = brownianMotions.T.dot(U)
        except Exception as e:
            msg = "Couldn't multiply uncorrelated Brownian increments with decomposed correlation matrix: %s, %s: %s" % (brownianMotions, U, e)
            raise DslError(msg)
        brownianMotionsCorrelated = brownianMotionsCorrelated.transpose()  # Put markets back on the first dimension.
        brownianMotionsDict = {}
        for i, marketName in enumerate(market_names):
            marketRvs = {}
            for j, fixingDate in enumerate(allDates):
                rv = brownianMotionsCorrelated[i][j]
                marketRvs[fixingDate] = rv
            brownianMotionsDict[marketName] = marketRvs

        return brownianMotionsDict