Hybrid computer system for asset selection optimization
A hybrid classical-quantum system addresses the inefficiencies of classical methods in portfolio selection by using a quantum circuit to optimize asset combinations, resulting in faster and more efficient computational solutions.
Patent Information
- Application Number
- PCT/CN2023/137935
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-12-11
- Publication Date
- 2025-06-19
AI Technical Summary
Classical computational methods for portfolio selection, such as Monte Carlo Simulation, are computationally intensive and inefficient in solving complex optimization problems.
A computer-implemented method using a hybrid classical-quantum system to perform asset selection optimization, involving the generation of a quantum circuit with modules for parameter loading, portfolio loading, computation, and quantum estimation to find the optimal asset combination that minimizes or maximizes a selection metric.
The method significantly reduces computational complexity and time by leveraging quantum computing to efficiently identify optimal asset combinations, thereby improving the efficiency of portfolio selection processes.
Smart Images

Figure CN2023137935_19062025_PF_FP_ABST
Abstract
Description
Hybrid Computer System For Asset Selection OptimizationFIELD OF THE INVENTION
[0001] The present application relates to methods and systems for providing efficient computational approaches to solving certain optimization problems involving combinations of assets in an asset portfolio.BACKGROUND OF THE INVENTION
[0002] Quantum computing has the potential to revolutionize data processing in the coming years due to the development of larger-qubit quantum computers. One of the significant applications of quantum computing across a range of engineering and business domains is the ability to efficiently solve complex optimization and machine learning problems.
[0003] Portfolio selection is a challenging problem in classical computation. Portfolio selection involves selecting a combination of assets of differing characteristics to meet some overriding selection criterion –e.g. to minimize or maximize some metric. Such assets may e.g. be financial assets (such as investment assets) . Monte Carlo Simulation is a widely used technique in various financial scenarios, including portfolio selection. The general idea behind Monte Carlo Simulation is to simulate potential portfolios and determine the optimal one based on the level of risk aversion.
[0004] However, the underlying techniques of selecting entities (assets) to combine in some way (e.g. to form a “portfolio” of assets) are more widely applicable than just financial applications. Such techniques can also be applied to other combinatorial optimization problems, including in the fields of computer vision, artificial intelligence and the like.
[0005] Purely classical implementations of optimization techniques (e.g. Monte Carlo simulation) can be very compute-intensive. Embodiments of the invention thus aim to provide more computationally efficient approaches based on utilizing quantum computing to solve portfolio selection problems.SUMMARY OF THE INVENTION
[0006] Aspects of the invention are set out in the independent claims. Certain preferred features are set out in the dependent claims.
[0007] Disclosed herein is a computer-implemented method for performing asset selection, comprising:
[0008] receiving asset data comprising asset parameters defining a plurality of assets;
[0009] generating control data defining a quantum circuit having a plurality of circuit modules, including:
[0010] a parameter loading module arranged to encode the asset parameters of the plurality of assets onto a set of qubits of a quantum computer;
[0011] a portfolio loading module arranged to encode a set of possible asset combinations as a superposition of a further set of qubits of the quantum computer, each asset combination corresponding to a selection of one or more of the plurality of assets;
[0012] a computation module for calculating a portfolio selection metric based on the asset parameters and the possible asset combinations, wherein the computation module is arranged to produce a quantum superposition state of the quantum computer that corresponds to values of the selection metric for each of the asset combinations encoded by the portfolio loading module; and
[0013] a quantum estimation module arranged to perform quantum estimation to find an optimal value in the superposition state that corresponds to an optimal value of the selection metric;
[0014] transmitting the control data defining the quantum circuit to the quantum computer for execution;
[0015] identifying from an output of the quantum computer the asset combination of the set of possible asset combinations that corresponds to the optimal selection metric value identified by the quantum estimation module; and
[0016] generating an output specifying the identified asset combination.
[0017] The method may further comprise determining the optimal value of the selection metric obtained by the quantum estimation module; and outputting the optimal selection metric value.
[0018] The method may comprise controlling the quantum computer to execute the quantum circuit and obtain one or more outputs of the quantum computation, the outputs preferably including the identifying asset combination (optionally including asset weights for each asset) and / or the optimal value of the selection metric. Execution of the quantum estimation module by the quantum computer preferably results in collapse of the superposition state of the quantum computer to a single state corresponding to the optimal value of the selection metric and / or to the asset combination giving rise that optimal value.
[0019] Preferably, outputs of the quantum computer specifying the identified asset combination and / or optimal value of the selection metric are obtained at the quantum computer by performing measurements of the associated qubits of the quantum computer after completion of the quantum estimation, preferably comprising: measuring the state of the further qubits encoding the set of possible asset combinations after collapse of the superposition state to identify the asset combination of the set of possible asset combinations that corresponds to the optimal selection metric value.
[0020] At least the receiving and generating steps, and preferably also the transmitting, identifying and / or generating steps, are performed by a classical computer system. The classical computer system may include the quantum computer or be connected to the quantum computer e.g. via a network.
[0021] The assets may be financial assets, and the asset parameters may comprise, for each asset: a measure of asset returns; and / or a measure of asset risk or volatility.
[0022] The optimal value preferably comprises a minimum or maximum value of the selection metric. In examples described herein, the method is used to minimize the selection metric (such as portfolio volatility) . However, the method may also be used to maximize a selection metric. In any case, as the skilled person understands, maximization problems can be recast as minimization problems and vice versa, and so minimization may be substituted by maximization or vice versa in the embodiments described herein.
[0023] In some examples, asset combinations are specified by a binary selection indicator associated with each asset specifying inclusion or exclusion of the asset in the combination. The binary selection indicators associated with the assets are preferably encoded by respective ones of the further set of qubits of the quantum computer.
[0024] Alternatively, asset combinations may be specified by a respective asset weight value associated with each asset indicating a quantity of the asset to be included in a portfolio of assets. Each asset weight is preferably encoded using a plurality of qubits of the further set of qubits of the quantum computer, allowing a range of weight values to be represented (note that a binary indicator may be considered a special case of a weight value, in which the weight can only take the values one or zero) . Asset weights associated with respective assets may specify (relative) proportions of each asset in the combination, for example as a percentage or fraction, e.g. in relation to an asset budget.
[0025] The method may comprise receiving configuration data defining the set of possible asset combinations to be evaluated, the configuration data optionally specifying a number of asset combinations and / or a number of bits to be used to encode each asset weight of an asset combination. The portfolio loading module of the quantum circuit may then be generated in dependence on the configuration data.
[0026] The selection metric may comprise a measure of portfolio risk or volatility. Preferably, the identified asset combination comprises a selection of one or more of the plurality of assets according to proportions specified by weight values of the particular asset combination that is associated with the lowest portfolio risk or volatility computed by the computation module. The selection metric may be computed as (or based on) a weighted sum of asset volatilities, weighted by the asset weights.
[0027] Executing the quantum circuit preferably comprises initializing the quantum state of the quantum computer to a superposition state that represents a plurality of possible asset combinations and their associated selection metric values. The quantum estimation module preferably performs operations for:
[0028] selecting a benchmark value of the selection metric;
[0029] repeating a process of:
[0030] amplifying marked states, wherein the marked states are states associated with a selection metric value that is lower than the benchmark value for minimisation of the selection metric or higher than the benchmark value for maximisation of the selection metric, and decreasing the amplitudes of other states of the superposition of states;
[0031] updating the benchmark to the lowest marked state for minimisation or the highest marked state for maximisation;
[0032] until at most one marked state remains; and
[0033] measuring the quantum state to obtain the minimum or maximum value of the selection metric and / or the associated asset combination.
[0034] Preferably, the amplification operations comprise one or more of: application of the oracle operator; application of the Grover operator.
[0035] The computation module preferably uses a weighted adder function to compute a weighted sum of the of the asset parameters according to the asset weights associated with the possible asset combinations, the selection metric computed based on the weighted sum.
[0036] Preferably, the output specifies a selected asset portfolio including: one or more assets of the plurality of assets included in the portfolio; and / or a quantity or proportion of each of one or more assets included in the portfolio, the quantity or proportion based on asset weights associated with the assets in the identified asset combination. The method may further comprise outputting the asset combination or selected asset portfolio to a user and / or initiating one or more trading transactions in accordance with the asset combination / selected asset portfolio, for example by transmitting control data to a transaction system.
[0037] Also disclosed herein is method for performing asset selection, the method performed using a quantum computer, the method comprising:
[0038] receiving asset data comprising asset parameters defining a plurality of assets;
[0039] encoding the asset parameters of the plurality of assets onto a set of qubits of the quantum computer;
[0040] encoding a set of possible asset combinations as a superposition of a further set of qubits of the quantum computer, each asset combination corresponding to a selection of one or more of the plurality of assets;
[0041] calculating a portfolio selection metric based on the asset parameters and the possible asset combinations, wherein the calculation is performed using quantum arithmetic operations to produce a quantum superposition state of the quantum computer that corresponds to values of the selection metric for each of the asset combinations;
[0042] performing quantum estimation to find an optimal value of the selection metric, wherein the quantum estimation process collapses the superposition state to a state that corresponds to the optimal value of the selection metric;
[0043] measuring the state of the further qubits encoding the set of possible asset combinations after collapse of the superposition state to identify the asset combination of the set of possible asset combinations that corresponds to the optimal selection metric value identified by the quantum estimation module; and
[0044] generating an output specifying the identified asset combination.
[0045] The method in this example may comprise the further steps or features of any method set out above or described in more detail below.
[0046] The disclosure also encompasses a system having means, optionally comprising a computer device having a processor with associated memory, for performing any method as set out herein. The system may further comprise an interface for communication with a quantum computer coupled to the computer device, or may comprise said quantum computer.
[0047] The disclosure also encompasses a computer program or computer readable medium comprising software code adapted, when executed by a data processing system, to perform any method as set out herein, and / or a computer readable medium comprising control data defining a quantum circuit for performing any quantum computation described herein.
[0048] Features of one aspect or example may be applied to other aspects or examples, in any combination. For example, method features may be applied to system or computer program aspects or examples (and vice versa) .
[0049] BRIEF DESCRIPTION OF THE FIGURES
[0050] Certain embodiments of the invention will now be described by way of example only, in relation to the Figures, wherein:
[0051] Figure 1A illustrates a general workflow that can be followed to solve the portfolio selection problem using quantum computing;
[0052] Figure 1B illustrates a process for generating a quantum circuit to perform portfolio selection;
[0053] Figure 2 illustrates in overview a quantum circuit that can be leveraged to solve the portfolio selection problem;
[0054] Figure 3 illustrates in overview a quantum circuit for performing quantum estimation;
[0055] Figures 4A-4B illustrate an example of portfolio loading that includes three assets and a budget of one, with each asset being able to be chosen or not;
[0056] Figures 5A-5B illustrate an example of portfolio loading that includes two assets and a budget of one, with each asset being able to have different weights;
[0057] Figures 6A-6B illustrate an example of volatility calculation using described techniques;
[0058] Figure 7 illustrates pseudo-code for implementing quantum minimum value finding; and
[0059] Figure 8 illustrates a hybrid classical / quantum computing system for implementing the described techniques.DETAILED DESCRIPTION
[0060] The following is a detailed description of exemplary embodiments to illustrate the principles of the invention. The embodiments are provided to illustrate aspects of the invention, but the invention is not limited to any embodiment. The scope of the invention encompasses numerous alternatives, modifications and equivalents; it is limited only by the claims.
[0061] Numerous specific details are set forth in the following description in order to provide a thorough understanding of the invention. However, the invention may be practiced according to the claims without some or all of these specific details. For the purpose of clarity, technical material that is known in the technical fields related to the invention has not been described in detail so that the invention is not unnecessarily obscured.
[0062] Embodiments of the invention provide a method for identifying a combination of assets from a set of available assets that minimizes a selection metric. In an embodiment, the assets may be financial assets, and a particular combination of assets is referred to as a portfolio.
[0063] The described techniques use a quantum computing approach for portfolio selection in order to efficiently identify better portfolios. The method can be used in a variety of portfolio selection cases. Furthermore, the described techniques can also be used in other scenarios where variable combinations of various assets can be selected from a set of available assets. For example, the techniques could be used to select a portfolio of server resources (as the “assets” ) for hosting a set of server workloads, e.g. based on server hardware resources, performance metrics, load statistics / projections and the like.
[0064] Embodiments exploit quantum superposition to efficiently find the desired portfolio. In a quantum system, the superposition of states can be represented by a superposition vector, which is a mathematical object that describes the probabilities of finding the system in each of its possible states. It allows quantum computers to perform certain calculations much faster than classical computers.
[0065] Figure 1 provides a visual representation of the workflow for solving portfolio selection for a portfolio of financial assets using quantum computing. The figure highlights the four key steps involved in the process, which are described below in greater detail.
[0066] Loading financial parameters
[0067] The first step in the workflow (shown as step 102 in Figure 1) involves loading financial parameters for the assets in the portfolio. These parameters are typically based on historical data concerning returns and volatilities of those assets. The historical data is used to estimate the risk and return characteristics of each asset. In an embodiment, volatility parameters for each asset are derived from historical asset data and are input into the quantum algorithm (encoded by quantum bits) to enable it to perform calculations and generate an optimal portfolio strategy. Volatility refers to the degree of fluctuation or variability in the price or value of a financial instrument, asset, or market over a specific period.
[0068] Loading Portfolio Strategies
[0069] In the second step (step 104) , different portfolio strategies are loaded onto the quantum circuit. This allows for the consideration of different combinations of assets within the portfolio.
[0070] Portfolio strategies are defined by a set of weights, one weight for each available asset, specifying a contribution of the asset to the portfolio. By varying the weights assigned to each asset, a range of different portfolios can be considered. Each weight may be represented using multiple bits, in which case the weight value indicates a relative proportion of the asset included in the portfolio. In some cases, the weight for each asset may be represented by a single bit, in which case the weight simply indicates inclusion or exclusion of that asset in the portfolio.
[0071] For representation on the quantum circuit, the weight for each asset is mapped to a set of qubits of the quantum computer and a superposition state is created representing different possible weight combinations.
[0072] Calculating the Selection Metric
[0073] Once the financial parameters and portfolio strategies are loaded onto the quantum circuit, the next step is to calculate the selection metric (step 106) . The selection metric (for example a measure of portfolio risk or volatility) is used to select a portfolio from the possible portfolio strategies. The metric is calculated using quantum arithmetic to obtain a superposition of values representing the desired value, such as the portfolio volatility. By utilizing quantum arithmetic, the proposed quantum native method can represent multiple values simultaneously and thus obtain results more efficiently. Present examples use portfolio volatility as the selection metric but other metrics could be used.
[0074] Using Quantum Estimation to Find the Minimum Value
[0075] The final step in the workflow involves using quantum estimation to find the minimum value in the superposition of the selection metric. This minimum value corresponds to the optimal portfolio strategy with the desired characteristics, as measured by the selection metric, in the present example the minimum portfolio volatility. By leveraging quantum estimation techniques, the proposed quantum native method can efficiently identify the optimal portfolio strategy for a given set of financial parameters.
[0076] Note that quantum estimation can equivalently be used to maximize the selection metric. The choice of selection metric and whether the metric is minimized or maximized will depend on the aims for portfolio selection. Examples below assume that a portfolio volatility metric is being minimized.
[0077] The above process is performed using a quantum circuit comprising sub-circuits (modules) for each of the described processing stages. Generation of the quantum circuit is summarised in Figure 1B.
[0078] In Step 120, the system generates the circuit module for loading the financial parameters.
[0079] Input: financial parameters, such as asset returns and / or volatilities.
[0080] Output: generated circuit for loading parameter values for each asset onto qubits of the quantum circuit
[0081] In step 122, the system generates a circuit module for loading the portfolio strategies.
[0082] Input: Data defining the portfolio strategies to be evaluated. The number of possible portfolio strategies is determined by the number of qubits used to represent the portfolio strategy as a set of asset weights, which is given by the number of available assets and the number of qubits used per asset weight. For example, for binary weights and n assets, 2n portfolio strategies can be encoded.
[0083] Output: generated circuits to simulate the number of portfolio strategies. The circuit generates the superposition state representing all possible strategies (as all possible weight combinations) .
[0084] In step 124, the system generates a circuit module for computing the selection metric.
[0085] Input: Operations on different variables. The operations define the mathematical expression that is used to compute the selection metric (e.g. portfolio volatility) from the asset parameters (e.g. asset volatility) and portfolio strategy (i.e. the weights determining the weighted asset composition of a given portfolio) . The specific set of operations is configurable. In an example, the selection metric may be the squared portfolio volatility, computed as the weighted sum of the squared asset volatilities, weighted according to the respective asset weights. The objective is to find the optimal portfolio strategy that satisfies this linear relationship and minimizes the portfolio volatility. In many cases, the exact value of the portfolio volatility may not be as important as achieving this objective.
[0086] More complex expressions could be used to compute the selection metric, for example incorporating additional asset parameters (e.g. asset volatilities and returns) .
[0087] Output: generated circuits to calculate the selection metric.
[0088] In step 126, the system combines the modules for loading financial parameters, loading portfolio strategies and computing the selection metric with the quantum estimation module.
[0089] Input: None.
[0090] Output: the overall circuit.
[0091] In step 128, the completed circuit is transmitted to the quantum computer for execution and the output of the circuit is obtained. The output of the circuit is the estimated expectation value produced by the quantum estimation algorithm, corresponding to the minimum value of the selection metric.
[0092] After applying the estimation algorithm, the minimum return state will be measured with the highest probability. The measurement of the return states also results in collapse of the quantum states of the qubits representing the asset weights in the portfolio strategy. The states of those qubits after the measurement thus indicate the asset weights that give rise to the minimum selection metric value. Thus, the final operation of the quantum circuit involves measuring the states of those qubits to obtain the final weight values corresponding to the selected portfolio strategy. Those weight values define the assets to be included in the final selected portfolio (and, for multi-bit weights, the proportions in which the assets are included in the portfolio) .
[0093] Figure 2 illustrates in overview a quantum circuit that can be used to solve the portfolio selection problem. The circuit consists of the four modules discussed, each with a specific function in the portfolio selection process. The four modules are:
[0094] Financial Parameters Loading 202: This module is responsible for loading the financial parameters for the assets in the portfolio onto the circuit. The financial parameters are encoded onto qubits, which are then used as input for the next module.
[0095] Portfolio Loading 204: This module is responsible for loading the portfolio strategies onto the circuit. The portfolio strategies are represented by the asset weights encoded onto a set of qubits (in a superposition state) , which are then used as input for the next module.
[0096] Risk Computation 206: This module is responsible for calculating the desired selection metric, such as the portfolio risk / volatility. This is done using quantum arithmetic to obtain a superposition state that includes all possible values of the volatility for the set of possible portfolios (defined by the possible combinations of weights for the assets) .
[0097] Quantum Estimation 208: This module is responsible for using quantum counting algorithms and estimation techniques to find the minimum value of the selection metric in the superposition state generated by the previous module. The minimum value represents the optimal portfolio strategy with the desired risk and return characteristics.
[0098] Figures 3-6 illustrate quantum circuit implementations for the individual modules in more detail. In the quantum circuits, the depicted circuit elements are defined as follows:
[0099] ● Rx gate: A rotation gate that rotates the qubit state around the y-axis on the Bloch sphere by a specified angle. It is often used to prepare superposition states and to perform single-qubit operations.
[0100] ● X gate: Also called the "Pauli-X" gate, it is a single-qubit gate that performs a bit-flip operation. It flips the state of the qubit from |0> to |1> or vice versa.
[0101] ● H gate: Also called the "Hadamard" gate, it is a single-qubit gate that prepares a superposition state by putting the qubit into an equal probability of being in the |0> or |1> state. It is often used to create entangled states and to perform quantum Fourier transforms.
[0102] ● CX gate: Also called the "CNOT" gate, it is a two-qubit gate that performs a conditional bit-flip operation on the target qubit if the control qubit is in the |1> state. It is often used to create entangled states and to perform quantum error correction.
[0103] ● M: a process that measures the state of a qubit. It collapses the superposition state of the qubit into a classical state, and the result of the measurement is a classical bit that can be used for further computation or communication.
[0104] Figure 3 illustrates a general circuit for implementing quantum amplitude estimation (which can be used to implement module 208 of Figure 2) , which is a known quantum algorithm for estimating the mean value of a set of states. The quantum amplitude estimation method consists of two operators: the oracle operator 302 and the Grover operator 304. These operators work together to rotate the quantum state and obtain the minimum value. The following is a detailed explanation of the different components of Figure 3:
[0105] Quantum State Initialization: At the beginning of the process, the quantum state is initialized to a superposition state that represents all possible values of the input parameters. In the present case the superposition of input parameters is created as the output of the risk computation module 206 (Figure 2) .
[0106] Oracle Operator 302: The oracle operator is designed to amplify the amplitude of a marked state, which is the state that satisfies a certain condition, and decrease the amplitudes of the other states. This is done by applying a phase flip to the marked state.
[0107] Grover Operator 304: The Grover operator is used to amplify the amplitude of the marked state further by rotating the quantum state in a specific way. The Grover operator consists of two steps: the first step is to apply a Hadamard transform to the quantum state, and the second step is to apply a phase flip to the state that is orthogonal to the marked state.
[0108] Amplitude Estimation: The amplitude estimation algorithm is used to estimate the mean value of the set of states. This is done by applying the oracle operator and the Grover operator repeatedly, which rotates the quantum state and amplifies the amplitude of the marked state. The number of iterations required to obtain a sufficiently accurate estimate depends on the desired level of precision and the number of input parameters.
[0109] Measurement 306: The final step in the process is to measure the quantum state and obtain the minimum value. Measurement is represented by the “M” elements in Figure 3. The measurement results in the collapse of the superposition state into a single state that corresponds to the minimum value of the selection metric.
[0110] As noted above, after collapse of the superposition state, the qubits representing the portfolio weights can then be measured to obtain the weights that give rise to the minimum value of the selection metric. These determine the “winning” portfolio strategy.
[0111] The quantum estimation algorithm will be described in more detail later.
[0112] Figure 4A provides an example of the portfolio loading process using binary weight variables to indicate the range of possible portfolio strategies. Figure 4A illustrates how the proposed quantum native method can be used to represent different portfolio strategies using superposition states, and how these states can be measured to obtain a portfolio corresponding to the minimum value of the selection metric. The following is a detailed explanation of the different components of Figure 4A:
[0113] Asset and Budget Allocation: The example in Figure 4A involves three assets. Each asset can either be chosen or not, which is represented by a qubit (q0 q1 q2) for each asset, resulting in eight possible portfolio combinations. The example further assumes a budget of one unit, meaning that only a single unit can be chosen, reducing the number of valid portfolio strategies to three –001, 010 and 100. For a different budget (e.g. two units) other portfolio combinations would become possible.
[0114] Quantum State Preparation: The binary variable representation of each portfolio combination is used to prepare the quantum state. The quantum state is a superposition of all possible portfolio combinations, i.e. all possible state combinations of the three qubits q0 q1 q2.
[0115] Measurement: As noted above, measurement of the output of the quantum estimation module results in the collapse of the superposition state into a single state that corresponds to one of the portfolio strategies. Thus, when the quantum state of the portfolio qubits is measured, one of the possible output states will be obtained (001, 010, or 100) . By way of illustration, Figure 4B illustrates an example of measured probabilities for each portfolio combination measured before the quantum estimation module collapses the superposition state.
[0116] Figure 5A provides an example of the portfolio loading process using multi-bit variables to represent integer-valued weights for each asset. Figure 5A illustrates how the proposed quantum native method can be used to represent different portfolio strategies using superposition states, which can then be used to perform calculations and find the portfolio corresponding to the minimum value of the selection metric. The following is a detailed explanation of the different components of Figure 5A:
[0117] Asset and Budget Allocation: The example in Figure 5A involves two assets with a budget of one, and each asset can have different integer weights. The weights are encoded using three qubits (q0-q2 for the first asset and q3-q5 for the second asset) , allowing 8 possible weight values for each asset. Each weight associated with a given asset indicates a proportion of the asset budget to be allocated to that asset, with a value zero for a weight indicating a proportion of 0%, i.e., the asset is not chosen, and a maximum value, here binary 111, indicating a proportion of 100%, i.e. the portfolio includes only that asset. Intermediate values represent respective proportions (percentages) of the portfolio allocated to that asset. Thus, different combinations of weight values result in a range of possible portfolio combinations. The percentages defined by each asset weight sum to 100%, representing the total asset budget.
[0118] Quantum State Preparation: The integer variable representation of the weights of each portfolio combination is used to prepare the quantum state. The quantum state is a superposition of all possible portfolio combinations (combinations of weight values for different assets) . Not all weight combinations may be valid -e.g. only those that add to 100%can be chosen. If necessary, restrictions to valid portfolio strategies can be enforced by the “portfolio loading” circuit.
[0119] Measurement: When the quantum state is measured, different states can be observed that represent different portfolio combinations. The weights for each asset determine the proportion of that asset to be included in the portfolio. For example, for three-bit weights (allowing for 8 different weight values including zero) , the proportion can be computed as wi / (23-1) . For example, w1=111, w2=000 means we only choose the first asset without any second asset, while w1=001, w2=110 represents a choice of 14.3%of the first asset and 85.7%of the second asset.
[0120] By way of illustration, Figure 5B again shows example measurement results, giving measured probabilities for each possible combination of weights for the two assets prior to operation of the quantum estimation module.
[0121] Additional assets may be represented by expanding the Figure 5A circuit using additional qubits. Furthermore, in the depicted example, thee qubits are used to represent weights for each asset, allowing for 23=8 possible weight values. However, a different number of qubits could be used to represent weights, for example to allow for more fine-grained weight increments.
[0122] Figure 6A provides an example of how the proposed portfolio selection method can be used to incorporate asset weights into the portfolio loading process and compute the volatility of the resulting portfolio options. The figure highlights how the proposed method can leverage Qiskit to add the financial parameters for each asset to the circuit and calculate the desired value, such as the portfolio volatility. The following is a detailed explanation of the different components of Figure 6A:
[0123] Asset and Budget Allocation: The example in Figure 6A involves two assets with a budget of one, and each asset is associated with a binary weight. This results in a range of possible portfolio combinations. The qubits q0 and q1 correspond to the binary weights.
[0124] Portfolio Loading (602) : The proposed quantum native method is used to load the portfolio strategies onto the circuit. The resulting portfolio options are represented by qubits and can be expressed as binary strings such as 01 or 10, indicating the presence or absence of each asset in the portfolio.
[0125] Weighted Adder Function (604, 606) : The weighted adder function from Qiskit is used to add the financial parameters for each asset to the circuit. The function computes the weighted sum of the financial parameters (e.g. asset volatilities or squared volatilities, encoded on qubits q2 and q3) for each asset, which can be used to calculate the desired value, such as the portfolio volatility. Note this is an illustrative example showing only a single qubit per asset parameter and in practice multiple qubits may be used to represent these values. Input values of (squared) volatilities for the assets are mapped by a suitable quantization to the available qubits.
[0126] Volatility Computation: The desired value, such as the portfolio volatility, is computed using quantum arithmetic to obtain a superposition state that includes all possible values of the volatility. In the present example, the squared volatility is computed as the weighted sum of squared asset volatilities (as discussed above) which is computed by the weighted adder functions. However, in other examples, different or additional computations could be performed. The resulting (squared) volatility values are post-processed (stage 608) with a linear transformation and are measured on qubits q4 and q5. A final volatility value can then be determined as where num_qubits is the number of qubits used to represent the squared volatility.
[0127] Figure 6B shows the resulting portfolio options in a bar graph, which provides a visual representation of the different portfolio strategies and their corresponding calculated volatilities (x-axis) with the corresponding measurement probabilities (y-axis) , prior to quantum estimation.
[0128] Figure 7 illustrates the quantum counting algorithm that the proposed method uses to find the minimum value of the superposition state. The figure highlights the key steps of the algorithm. The main steps are as follows:
[0129] Initial Guess: The algorithm begins with an initial guess for the selection metric, which serves as a benchmark for the quantum counting algorithm. The initial guess can be any value within the range of possible values.
[0130] Quantum Counting Algorithm: The quantum counting algorithm is used to search for the number of states that are lower than the benchmark. This is done by applying the Grover operator, which amplifies the amplitude of the marked state that satisfies the condition of being lower than the benchmark. The algorithm is repeated multiple times until the number of marked states is found.
[0131] Benchmark Update: If the number of marked states is not equal to 0, the benchmark is updated to the lowest marked state. This ensures that the new benchmark is less than or equal to all marked states (note in the case of maximisation rather than minimisation this process would be reversed, with the benchmark updated to the highest marked state to ensure the new benchmark is greater than or equal to all marked states) .
[0132] Minimum Value Finding: When the number of marked states is equal to 1, the minimum value of the selection metric can be measured with 100 percent probability. This results in the collapse of the superposition state into a single state that corresponds to the minimum value of the selection metric. As a result, the state of qubits representing the portfolio weights will also have collapsed, so by measuring those qubits, it is possible to read the portfolio weights corresponding to the lowest selection metric (e.g. lowest volatility) .
[0133] Once the portfolio strategy associated with the minimum value of the selection metric has been identified, it can be output to the user. For binary portfolio selection, this may simply indicate which of the available assets to include / not include in the portfolio. For weighted selection, this may indicate the relative proportions (e.g. in percentages) of each asset to include in the portfolio, as determined by the weights of the identified portfolio strategy with the minimum selection metric value (e.g. specifying 14.3%of a first asset type and 85.7%of a second asset type as in the above example for weight values of w1=001, w2=110) .
[0134] Rather than merely outputting relative proportions, the system may additionally compute actual asset quantities to be purchased, e.g. by applying the proportions to a real-world spending budget. The system may additionally configure and / or perform automatic transactions to purchase the identified combination of assets in accordance with the identified portfolio strategy (e.g. weights) using an asset trading computer system.
[0135] Example computer system
[0136] Figure 8 illustrates a hybrid system comprising classical and quantum computing devices for implementing described techniques.
[0137] The processing device 800 is a classical computing device and may be based on conventional workstation or server hardware. Thus, processing device 800 includes one or more processors 804 together with volatile / random access memory 802 for storing temporary data and software code being executed. Local I / O interfaces 806 provide interfaces to local devices (e.g. keyboards / displays) to enable a user to interact with the system. Alternatively, user interaction may be remote (e.g. in a server implementation) .
[0138] Persistent storage 810 (e.g. in the form of hard disk storage, optical storage and the like) persistently stores software and data for performing various described functions. In an example, this includes asset and portfolio data 812 defining assets and portfolio strategies (which may e.g. by configured by a user) and a control process 814 for generating the quantum circuit implementing the portfolio selection process and controlling execution of the circuit on the quantum computing system 840.
[0139] The persistent storage further includes a computer operating system 816 and any other software and data needed for operating the processing device. The device will include other conventional hardware components as known to those skilled in the art, and the components are interconnected by one or more data buses (e.g. a memory bus and I / O bus) .
[0140] A network interface 820 is provided for communication with other system components. For example, the processing device may communicate via the network interface with the external quantum computing system 840 (to run quantum circuits on the quantum processing unit 844 via the quantum controller 842) . Communication may occur over one or more networks 822 (e.g. Local and / or Wide Area Networks, including private networks and / or public networks such as the Internet) . The quantum computing system may be provided as a cloud-based quantum computing service accessible via the internet. Alternatively, the quantum computing system could be a local quantum computer connected to the processing device 800 via the network interface or via a wired / wireless peripheral interface (such as USB) etc. or could be integrated into the processing device 800. The quantum computer may be any suitable quantum computer, such as an IBM Osprey or other IBM quantum computer, a Rigetti Aspen M-2 system etc.
[0141] Quantum circuits define operations (e.g. in the form of quantum gates) to be performed by the quantum computing system and thus constitute control data for controlling operation of the quantum computer. The circuits may e.g. be expressed using a quantum programming language such as QASM (or variations of that language) . Quantum circuits may then be compiled to generate control signals to be applied to the quantum processing unit, e.g. by quantum controller 842.
[0142] While a specific architecture is shown and described by way of example, any appropriate hardware / software architecture may be employed to implement the hybrid computer system.
[0143] Furthermore, functional components indicated as separate may be combined and vice versa. For example, the functions of the system may be performed by a single device 800 or may be distributed across multiple devices.
[0144] It will be understood that the present invention has been described above purely by way of example, and modification of detail can be made within the scope of the invention.
Claims
1.A computer-implemented method for performing asset selection, comprising:receiving asset data comprising asset parameters defining a plurality of assets;generating control data defining a quantum circuit having a plurality of circuit modules, including:a parameter loading module arranged to encode the asset parameters of the plurality of assets onto a set of qubits of a quantum computer;a portfolio loading module arranged to encode a set of possible asset combinations as a superposition of a further set of qubits of the quantum computer, each asset combination corresponding to a selection of one or more of the plurality of assets;a computation module for calculating a portfolio selection metric based on the asset parameters and the possible asset combinations, wherein the computation module is arranged to produce a quantum superposition state of the quantum computer that corresponds to values of the selection metric for each of the asset combinations encoded by the portfolio loading module; anda quantum estimation module arranged to perform quantum estimation to find an optimal value in the superposition state that corresponds to an optimal value of the selection metric;transmitting the control data defining the quantum circuit to the quantum computer for execution;identifying from an output of the quantum computer the asset combination of the set of possible asset combinations that corresponds to the optimal selection metric value identified by the quantum estimation module; andgenerating an output specifying the identified asset combination.2.A method according to claim 1, further comprising determining the optimal value of the selection metric obtained by the quantum estimation module; and outputting the optimal selection metric value.3.A method according to claim 1 or 2, wherein execution of the quantum estimation module results in collapse of the superposition state of the quantum computer to a single state corresponding to the optimal value of the selection metric and / or to the asset combination giving rise that optimal value.4.A method according to any of the preceding claims, wherein outputs of the quantum computer specifying the identified asset combination and / or optimal value of the selection metric are obtained at the quantum computer performing measurements of the associated qubits of the quantum computer after completion of the quantum estimation, preferably comprising:measuring the state of the further qubits encoding the set of possible asset combinations after collapse of the superposition state to identify the asset combination of the set of possible asset combinations that corresponds to the optimal selection metric value.5.A method according to any of the preceding claims, wherein at least the receiving and generating steps are performed by a classical computer system.6.A method according to any of the preceding claims, wherein the assets are financial assets, and wherein the asset parameters comprise, for each asset:a measure of asset returns; and / ora measure of asset risk or volatility.7.A method according to any of the preceding claims, wherein the optimal value comprises a minimum or maximum value of the selection metric.8.A method according to any of the preceding claims, wherein asset combinations are specified by a binary selection indicator associated with each asset specifying inclusion or exclusion of the asset in the combination.9.A method according to claim 8, wherein the binary selection indicators associated with the assets are encoded by respective ones of the further set of qubits of the quantum computer.10.A method according to any of claims 1 to 7, wherein asset combinations are specified by a respective asset weight value associated with each asset indicating a quantity of the asset to be included in a portfolio of assets.11.A method according to claim 10, wherein each asset weight is encoded using a plurality of qubits of the further set of qubits of the quantum computer.12.A method according to claim 10 or 11, wherein asset weights associated with respective assets specify relative proportions of each asset in the combination, optionally in relation to an asset budget.13.A method according to any of the preceding claims, comprising receiving configuration data defining the set of possible asset combinations to be evaluated, the configuration data optionally specifying a number of asset combinations and / or a number of bits to be used to encode each asset weight of an asset combination.14.A method according to any of the preceding claims, wherein the selection metric comprises a measure of portfolio risk or volatility.15.A method according to claim 14, wherein the identified asset combination comprises a selection of one or more of the plurality of assets according to proportions specified by weight values of the particular asset combination that is associated with the lowest portfolio risk or volatility computed by the computation module.16.A method according to any of the preceding claims, wherein executing the quantum circuit comprises initializing the quantum state of the quantum computer to a superposition state that represents a plurality of possible asset combinations and their associated selection metric values.17.A method according to claim 16, wherein the quantum estimation module performs operations for:selecting a benchmark value of the selection metric;repeating a process of:amplifying marked states, wherein the marked states are states associated with a selection metric value that is lower than the benchmark value for minimisation of the selection metric or higher than the benchmark value for maximisation of the selection metric, and decreasing the amplitudes of other states of the superposition of states;updating the benchmark to the lowest marked state for minimisation or the highest marked state for maximisation;until at most one marked state remains; andmeasuring the quantum state to obtain the minimum or maximum value of the selection metric and / or the associated asset combination.18.A method according to claim 17, wherein the amplification operations comprise one or more of: application of the oracle operator; application of the Grover operator.19.A method according to any of the preceding claims, wherein the computation module uses a weighted adder function to compute a weighted sum of the of the asset parameters according to the asset weights associated with the possible asset combinations, the selection metric computed based on the weighted sum.20.A method according to any of the preceding claims, wherein the output specifies a selected asset portfolio including:one or more assets of the plurality of assets included in the portfolio; and / ora quantity or proportion of each of one or more assets included in the portfolio, the quantity or proportion based on asset weights associated with the assets in the identified asset combination.21.A method for performing asset selection, the method performed using a quantum computer, the method comprising:receiving asset data comprising asset parameters defining a plurality of assets;encoding the asset parameters of the plurality of assets onto a set of qubits of the quantum computer;encoding a set of possible asset combinations as a superposition of a further set of qubits of the quantum computer, each asset combination corresponding to a selection of one or more of the plurality of assets;calculating a portfolio selection metric based on the asset parameters and the possible asset combinations, wherein the calculation is performed using quantum arithmetic operations to produce a quantum superposition state of the quantum computer that corresponds to values of the selection metric for each of the asset combinations;performing quantum estimation to find an optimal value of the selection metric, wherein the quantum estimation process collapses the superposition state to a state that corresponds to the optimal value of the selection metric;measuring the state of the further qubits encoding the set of possible asset combinations after collapse of the superposition state to identify the asset combination of the set of possible asset combinations that corresponds to the optimal selection metric value identified by the quantum estimation module; andgenerating an output specifying the identified asset combination.22.A method according to claim 21, further comprising the further steps or features of any of claims 2 to 20.23.A system having means, optionally comprising a computer device having a processor with associated memory, for performing a method according to any of the preceding claims.24.A system according to claim 23, further comprising an interface for communication with a quantum computer coupled to the computer device, or comprising said quantum computer.25.A computer program or computer readable medium comprising software code adapted, when executed by a data processing system, to perform a method as set out in any of claims 1 to 22.
Citation Information
Patent Citations
Optimal portfolio determination method based on variable component sub-line and related device
CN115545947A
Hierarchical portfolio optimization using clustering and near-term quantum computers
US20210133881A1
Systems and methods for quantum based optimization of a personalized portfolio
US20210374585A1
Systems and methods for quantum computing-assisted portfolio selection
US20230298101A1