Money Market Fund Risk Mitigation Path Analysis via Entropy Weight Method

Money market funds play an important role in the financial system as important means of short-term liquidity management and preservation of cash. In order to provide high security, fast liquidity and stable returns to investors, these funds mainly invest in short-term, high-quality debt instruments such as treasury bills, commercial paper and certificates of deposit ^[13]. But with the change of macroeconomic situation and increasing market volatility, these funds have been subject to many kinds of risks, such as credit risk, liquidity risk, interest rate risk, etc. There is a contradiction in the structure of money market funds; although their net asset value remains stable, the underlying assets are volatile and therefore at risk of fire sale by sudden redemptions in times of market stress ^[7]. Therefore, building a sound and objective risk mitigation system is necessary for the stability of the financial system and to protect the legitimate rights and interests of investors.

Traditional risk assessment methods are generally based on subjective weighting and cannot reflect the actual situation of various risk factors, thus resulting in the formulation of inappropriate risk reduction plans. To solve the problem of this method, the Entropy Weight Method is a rigorous and objective mathematical way to evaluate risk. Rooted in information theory, the basic idea of the Entropy Weight Method is that the dispersion of observed data for a given index is positively related to its information entropy; lower entropy means more information utility and thus a higher objective weight in the evaluation model. By calculating the entropy weight of each risk index, this method can avoid the subjectivity of artificial赋权 and obtain an objective risk profile that truly reflects the market ^[1].

In this paper, we will use the Entropy Weight Method to construct a standard operating procedure for the entire risk management of money market funds. The path of implementation will have three parts: first, create a multi-dimensional risk evaluation index system for credit risk, liquidity risk and market risk; second, based on the data fluctuation of each risk indicator, use the entropy weight to calculate its objective weight; third, combine the weighted indicators to form a risk mitigation plan. Based on the objective weight of the risk factors, the fund manager can assign risk buffers and adjust the portfolio allocation. The practical application value of this method is that it can turn the theoretical risk indicators into operation rules to improve the robustness of the money market fund and meet the high standard of regulatory liquidity in adverse market conditions. During periods of severe market stress, large-scale redemptions from money market funds can trigger destabilizing run dynamics that threaten broader market functioning ^[16], and large institutional investors typically do not differentiate their redemption decisions based on the liquidity positions of the funds they hold, which can amplify aggregate liquidity shortages ^[10]. Even with the deployment of central bank emergency liquidity tools to stabilize the sector, targeted regulatory reforms are still required to reduce inherent redemption incentives and curb systemic挤兑 risk ^[4].

In the environment of money market funds, risk appears more prominent than that of other securities investment funds because they are required to safeguard the principal and have high liquidity. Based on the general analysis method of the financial sector, this part will identify three main types of risks. Liquidity risk is the risk that the fund cannot fulfil a large-scale or sudden redemption request at net asset value and has to liquidate assets; it is mainly caused by a mismatch in the maturity of the portfolio's assets and liabilities. Credit risk means that there is a possibility of the short-term debt instruments in the portfolio defaulting or being downgraded, thus causing a loss of principal and interest payments that are late. Interest rate risk is the extent to which changes in market interest rates affect the price of fixed-income securities and, therefore, the fund's yield to maturity and market value.

The basic theories for the above risks are modern portfolio theory, liquidity preference theory, and asset-liability matching theory. Modern portfolio theory is used to reduce risk by diversifying investment and thus achieve a good risk-return relationship for short-term portfolios. Liquidity preference theory is used to explain why investors will demand a premium for less liquid, longer-maturity instruments and thus connect credit and interest rate risks with liquidity constraints. Asset-liability matching theory is the foundation for operation, so the duration and cash flow characteristics of the fund's assets must be in line with the expected redemption liabilities to mitigate liquidity risks.

There are deep-seated logical reasons for these risks in all stages of the entire operation, from issuance, daily operation to redemption settlement. In issuance, to pursue high returns, one may buy low-credit, long-maturity assets and thus increase the base credit and interest rate risk. With the increase in market interest rates in recent days, the value of assets has fallen, and thus interest rate risk will be increased; mark-to-market depreciation directly increases credit risk, and issuers face refinancing pressure, which can lead to severe liquidity risk if there are large-scale redemptions by institutional investors at the time of settlement. Therefore, this connectedness of risks requires the establishment of a holistic risk control system.

Based on all the risk factors above, we will use objective data processing. Subjective weighting cannot handle the change in risk of money market funds. So, the entropy weight method is quite good; it only considers the information entropy of the data of each risk indicator and will not be affected by people's subjectivity. The entropy weight method can improve the indicators with larger dispersion in the operation of money market funds according to the objective variance. In addition, according to the above method, the various forms of risk that do not meet the short-term debt nature of these funds (such as the long-term equity price fluctuation) are excluded, and a solid theoretical basis is provided for subsequent quantitative modelling and path planning.

Constructing the money market fund risk evaluation index system is the first condition to objectively quantify risk exposure and build a targeted risk mitigation path. Guided by the principles of data availability, representativeness, and relative independence, we have chosen 18 secondary indicators as candidates in three major dimensions: liquidity, credit, and interest rate at the beginning. This kind of multi-faceted system conforms to the operating characteristics of money market funds and thus emphasises capital preservation and high liquidity. To remove redundant information and improve the efficiency of statistics, correlation analysis was performed using five years' worth of daily and monthly operation data for 242 public offering money market funds. After this strict selection, variables with high correlations have been removed, and a refined set of key indicators has been obtained. The selected liquidity indicators are the proportion of assets with a maturity of less than seven days and the net redemption ratio on a single trading day; the credit dimension is indicated by the proportion of assets with an AAA credit rating, and the interest rate dimension is composed of the average duration gap and the 7-day annualised yield fluctuation range.

Each of the retained indicators has its own economic implication. The proportion of assets maturing within seven days can show whether the fund is able to meet the demand for immediate redemptions, and thus it is one of the indicators of liquidity resilience. The net redemption ratio in a single trading day can reflect the extent of investor panic and thus be used to measure behavioural liquidity pressure. The proportion of underlying assets with an AAA credit rating is the degree of default risk exposure of the portfolio, and thus it can indicate the safety margin of the investment. The average duration gap is used to analyse how sensitive the net asset value of a fund is to interest rate changes and therefore is a structural interest rate risk measure. Finally, the 7-day annualised yield fluctuation range is a measure of yield volatility and an indicator of market risk and operational instability.

In order to compare the different indicators uniformly, standard data processing rules will be established. First, for positive indicators such as the proportion of AAA-rated assets, we will apply normalization to unify the scale. Negative indicators such as the net redemption ratio and yield fluctuation amplitude are inversely transformed; thus, higher values are accompanied by lower risk. Interval indicators are those with a maturity of less than 7 days. Process all of them with the optimal range function, because too much concentration in short-term assets can lead to lower returns and be an indirect sign of stress. In order to keep the statistical properties of the fund's operation data, missing values will be filled with the moving average of the previous period, and abnormal values more than three standard deviations from the mean will be adjusted by winsorisation. The standardised preprocessing will ensure that the entropy weight calculation performed later is robust and reliable.

Entropy weight method is a kind of objective weight distribution technology that can be used to obtain the weight of indicators according to the information entropy of the actual data distribution. Information entropy is a measure of how disorganised and uncertain the data is; the higher the information entropy, the more uniform the distribution of data will be, and thus the less discriminating information it will provide, so a lower weight should be assigned to this indicator. In the opposite case, there will be more fluctuations and more useful information, so a relatively large weight will be assigned to it. Based on the results of data analysis, it avoids the subjectivity of traditional risk assessment models that depend on experts' judgments and reflects the risk situation more objectively.

In accordance with this, we will build a standardised calculation process for the risk evaluation index system of money market funds. First, we standardised the dimensionality of the preprocessed panel data for 242 sample funds from 2018 to 2023. Since risk indicators have different magnitudes and directions, min-max standardisation is used to standardise all data to an interval of [0,1]. Second, calculate the proportion of each fund's value in a certain index. Then calculate the information entropy of each indicator. Based on the entropy, subtract the entropy from one to obtain the information utility value, which is the effective information of the indicator used to evaluate the risk. Finally, calculate the objective weight of each secondary indicator by normalising the information utility values of all indicators. These secondary weights are then added together to determine the weight of each primary risk dimension.

The final weight distribution is quite different from the subjective weights that are often used in internal risk control systems for funds. Internal models usually overestimate credit risk and underestimate liquidity risk based on historical default assumptions. But entropy-weighted results give greater weight to liquidity and concentration risk indicators. The reason for the difference is that there are practical economic reasons between 2018 and 2023. Under market stress events, high-risk money market fund samples had severe redemption pressures and asset illiquidity; there were few real defaults. Objective data distribution shows that liquidity shortage and portfolio concentration are the reasons for the system vulnerability of money market funds; hence, objective risk measurement is needed.

Based on the quantitative results of the entropy weight method, this paper will systematically identify the main limiting factors that restrict the reduction of risks in money market funds. Based on the objective information entropy of the sample data, the basic idea of this identification is to select indicators with strong discrimination and avoid the arbitrariness of traditional risk factor selection. First, all evaluation indicators should be arranged in descending order based on their objective weight values. To set a threshold of 8% for the weight of an individual index, 7 indicators are selected to form the core constraint set. They are the basic characteristics that have the greatest impact on the risk level of money market funds and the most constraints.

Afterwards, in order to express these quantitative weights in a practical way, the identified constraint set is crossed-checked with the typical case data of 19 risk event funds in the sample period. The results of empirical research show that there are structural problems in the existing risk mitigation mechanism that have been overlooked for a long time. First, the speed of adjustment to changes in liquidity for assets does not match the rhythm of sudden large-scale redemptions. When there is widespread panic in the market, people rush to withdraw their money en masse, and the current system cannot liquidate all assets quickly enough to meet these demands without violating regulatory limits on liquidity, so a liquidity spiral will occur. Another problem is that insufficient risk identification depth for low-credit rating underlying assets has led to the classification of many of these assets as high-credit ones. Traditional mitigation is often based on superficial external ratings and does not consider the worsening of the basic credit situation, leading to unexpected default losses that reduce the fund's net asset value. In addition, at times when interest rates on the market change quickly, the asset portfolio does not adjust quickly enough in response to changes in interest rates. Conventional duration matching strategies are too slow to respond to changes in the yield curve; thus, they will suffer from large mark-to-market losses and fail to mitigate risks effectively.

Finally, to determine whether the identified core restrictive factors are robust and to rule out the possibility of single-sample disturbance, grouping regression tests are carried out for different fund sizes and different investment styles. The consistency of the regression coefficients shows that the three structural defects of liquidity-redeeming mismatch, credit risk misidentification, and interest rate adjustment lag remain universal restrictive factors regardless of scale or style changes, thus confirming their basic role in constructing targeted risk mitigation paths.

The dynamic liquidity risk matching system is essentially to resolve the problem of delayed adjustment and a lack of flexibility in the static Reserve Ratio of the old system; thus, we can adjust asset liquidity flexibly according to the changes in demand for payment. Based on the primary constraint identified by the entropy weight method with the highest weight, the aim of this mechanism is to strike a balance dynamically: continuously adjust the fund's maturity structure and liquidity buffer in response to changes in market trading sentiment and actual redemption demand. It is more practical to actively manage risks instead of passively defending them; otherwise, there will be a problem of inadequate liquidity and insufficient earnings.

First of all, the rules for dynamic adjustment of the asset maturity structure will be set. The system automatically responds to the real-time net redemption ratio and early warning signals of large-scale redemption, adjusting the portfolio allocation from long-term instruments to short-term ones as the redemption pressure increases. At the same time, a layered liquidity reserve pool of ultra-short-term treasury bonds and high-liquidity interbank certificates of deposit has been established by the system. Different risk response priorities are given to these instruments; ultra-short-term Treasury bonds are regarded as the first tier because they have a good market depth and stable valuation, and interbank certificates of deposit are chosen as the second tier to offer additional liquidity with slightly higher returns.

For the operation of this layered pool, a certain number of triggers for each level of liquidity response need to be determined. Based on the statistical situation of normal and abnormal redemption volumes, the threshold will be set according to the historical redemption data of the fund in the same period. When the real-time net redemption ratio exceeds these historical thresholds, the corresponding level of liquidity reserve will be activated one by one.

There is a standard operation process for adjusting the portfolio duration and handling non-current assets according to various warning levels. At the level of moderate alert, the proportional reduction of the portfolio's duration is required, and new investment in long-term assets will be suspended temporarily while the old ones are gradually retired. At the severe warning level, an immediate liquidation of the secondary reserve and a timely disposal of the non-current assets will be carried out according to the pre-established disposal plan, and the standardised valuation discount will be used to quickly convert the assets into cash. Based on data, this kind of systematic and data-driven solution can solve the problem that the static liquidity reserve ratio rule of the original mechanism cannot adapt to the changes in market trading sentiment for a long time, and it also provides a solid, verifiable way to deal with the liquidity risk of money market funds.

Based on the main reason for the second-largest weight in the entropy weight system, we propose a new credit risk tiering system to address the problem of default losses in money market funds. Fundamentally, this structure is to divide the underlying investable assets into several credit tiers according to various quantitative indicators; therefore, the degree of concentration of investment in various areas will be adjusted accordingly. This method does not take the risk exposure of all the assets into account; rather, the risk capital will be distributed according to the credit quality of the assets.

First, divide the investable assets into four different credit tiers. The grade is based on the comprehensive analysis of the issuer's credit rating, past default history, the operation status of the industry, and the degree of security provided by the bonds. Every level is increasing in risk; that is to say, the first type of assets has good collateral and a good record of repayment in the past; as the level goes down, the default rate of the assets increases and the basis of the industry becomes weaker. In order to avoid too much risk accumulation, the framework has set the upper limit of assets that money market funds can hold for each credit tier. The higher-level ones will be given larger allocation ranges, and the lowest level will have stricter concentration limits to limit the fund's exposure to severe credit problems.

A critical part of this framework's operation is to perform regular credit re-evaluation. The old method of regular credit rating tracking is relatively old and cannot catch up with the changes in the financial situation of issuers quickly. Therefore, the framework requires that the credit assessment be updated at least once in 10 trading days. Therefore, timely detection of early signs of credit distress can be carried out; thus, there is more time to prevent losses due to credit risk. In order to address the problem of relying on a single analysis, we will now introduce cross-checking rules for different sources of credit risk data. Rules that combine an internal quantitative model, a market-implied spread and other data sources can avoid excessive reliance on the rating results of one particular rating agency and cause deviations in the credit assessment.

At last, the system will take measures to reduce the positions of assets that are not very good in the lowest tier. When an asset falls below the lower level after being re-evaluated, a mandatory liquidation plan will be activated, and the fund will dispose of the deteriorating asset within a certain time period. In practice, this tiered system changes the static credit constraints into a dynamic, data-verified defence, improving the resistance of money market funds to extreme credit events and optimising the overall risk-adjusted return.

In the entropy weight analysis of restrictive factors, interest rate risk is the first to be identified; it is necessary to provide formal, executable hedging operation standards to reduce portfolio sensitivity. The definition of the aforementioned standard is to adjust the duration of the asset portfolio in line with changes in the liability benchmark to reduce the change in net value due to the change in the yield curve. The purpose is to adjust the portfolio's duration according to the trend of the market through standardised derivatives to convert the uncertain interest rate risk into a certain amount of change in basis points.

Real-time duration gap monitoring model should be built in real time according to changes in the market 7-day interbank offered rate and 1-year national bond yield. The two types of indicators are the present-day financing cost and medium-term income anchor of the fund, respectively. The model will continuously calculate the current duration gap and judge whether it is in the reasonable range by default. The range has been divided according to different levels of fluctuation in the market interest rate; in other words, the interval for a low-volatility range is larger, and for a high-volatility range, the permitted interval for duration is smaller.

Based on the above rules of the industry, the tools for standardized hedging operations are interest rate swaps and short-term interest rate futures. Interest rate swaps can convert floating-rate risks into fixed rates, and short-term interest rate futures can accurately adjust the liquidity of the duration. To guarantee compliance and liquidity safety, the maximum proportion of hedging positions to the total net asset value of the fund will be stipulated, and the exposure amount of derivatives will be controlled.

Based on the fluctuation range of market interest rates, the position of the hedging portfolio will be adjusted. If the 7-day interbank offered rate or the 1-year national bond yield is too high or low, a monitoring model will be activated to adjust. The operating instructions specify that the notional amount of swaps be increased or futures positions be adjusted to bring down the term gap to the allowed range. This rule-based, automatic adjustment is not a discretionary one, thus reducing the risk of changes in the net value of the asset portfolio due to a sudden change in market interest rate. To put this into practice through an institution, funds will have a strong and compliant way to cope with interest rate fluctuations.

In this paper, we will use the entropy weight method to build a risk reduction path for money market funds and some important information for theoretical improvement and actual investment management will be obtained. The basic idea of the entropy weight method comes from the theory of information entropy; it can objectively measure how dispersed the evaluation index is. The weight of the indicator is assigned in light of how large it is to avoid the problem of excessive subjectivity in the classic expert scoring method; thus, it can be used to assess the risk factors of liquidity, credit and interest rate risks quantitatively. The basic idea of this analysis is that objective risk weighting is required for formulating targeted and effective risk-reduction measures. If we do not know which particular risks have had the greatest impact on the fund's net asset value and caused the most significant fluctuations, all the later risk-reduction measures will be wrong.

The operational process to be applied in this study will follow a structured path. Firstly, based on the regulations and industry standards, an initial index system with multi-faceted risk indicators will be created. Secondly, data standardization will be carried out to reduce the dimensionality differences among various financial indicators. Thirdly, the information entropy value of each risk variable will be calculated to determine its objective weight. Finally, the general risk assessment score will be obtained and then used to guide the planning of specific measures to reduce risks. The process is also necessary in actual use. With the changes in the short-term financing market that money market funds face, there is a problem of asset maturity mismatch and sudden redemption pressure that may lead to a systemic liquidity shortage ^[3]. The entropy weight method can help the fund manager determine the most volatile risk point at any given time, shifting risk management from a passive compliance model to an active one and using more quantified data ^[15].

The specific application value of this method is that it can be directly used to convert the abstract risk calculation into practical measures to reduce risks. Since the dispersion of the liquid risk index is relatively high, the reduction path for this reason will be to reallocate funds to high-quality, short-term instruments and increase the liquidity reserve ^[6]. If the spread of credit risk increases, we should adopt the opposite path and strengthen the selection of counterparties and collateral ^[12]. Finally, based on this, it can be said that adding the entropy weight method to the risk management system of money market funds can provide a standardised, replicable and objective operation plan. It gives financial professionals a strong analytical tool to cope with market fluctuations, improve the match between assets and liabilities, and better protect investors' funds ^[9].

[1]朱丛倩. 互联网货币市场基金风险研究[D]. 浙江大学, 2018.

[2]邵莹, 项华彬. 我国货币市场基金风险问题研究[J]. 西部财会, 2018(2): 4.

[3]会计. 货币市场基金会计计量问题研究--以A货币市场基金为例[D]. 2019.

[4]邹子昂, 王淳. 美国货币市场基金的危机应对[J]. 中国金融, 2022(16): 82-84.

[5]Patrick E. McCabe. The Cross Section of Money Market Fund Risks and Financial Crises[J]. Finance and Economics Discussion Series, 2010.

[6]Philip E. Strahan, Başak Tanyeri. Once Burned, Twice Shy: Money Market Fund Responses to a Systemic Liquidity Shock[J]. Journal of Financial and Quantitative Analysis, 2015.

[7]Ramin Baghai, Mariassunta Giannetti, Ivika Jäger. Liability Structure and Risk Taking: Evidence from the Money Market Fund Industry[J]. Journal of Financial and Quantitative Analysis, 2017.

[8]Antoine Bouveret, Antoine Martin, Patrick E. McCabe. Money Market Fund Vulnerabilities: A Global Perspective[J]. Finance and Economics Discussion Series, 2022.

[9]Marcin T. Kacperczyk, Philipp Schnabl. How Safe Are Money Market Funds?*[J]. The Quarterly Journal of Economics, 2013.

[10]Fernando Avalos, Dora Xia. Investor size, liquidity and prime money market fund stress[J]. RePEc: Research Papers in Economics, 2021.

[11]Richard Makadok, Gordon Walker. Search and selection in the money market fund industry[J]. Strategic Management Journal, 2007.

[12]Emily Gallagher, Lawrence Schmidt, Allan Timmermann. Investor Information Acquisition and Money Market Fund Risk Rebalancing during the 2011–2012 Eurozone Crisis*[J]. Review of Financial Studies, 2019.

[13]Dale L. Domian, William Reichenstein. Performance and Persistence in Money Market Fund Returns[J]. Financial Services Review, 1997.

[14]Hilary J. Allen. Money Market Fund Reform Viewed Through A Systemic Risk Lens[J]. eYLS (Yale Law School), 2010.

[15]Ramon Paul DeGennaro, Dale L. Domian. Market Efficiency and Money Market Fund Portfolio Managers: Beliefs Versus Reality[J]. Financial Review, 1996.

[16]Lawrence Schmidt, Allan Timmermann, Russ R. Wermers. Runs on Money Market Mutual Funds[J]. American Economic Review, 2016.