Heterogeneous Trader Responses to Macroeconomic Surprises: A Simulation of Order‑Flow Dynamics

A consistent finding in recent empirical work is that major macroeconomic announcements provoke significant trading activity in financial markets. For example, an extensive study of UK markets by Heinlein and Lepori finds that a “large number of macroeconomic announcements increase trading activity in the stock market,” indicating that news releases lead market participants to revise expectations and trade accordingly. And the same research mentions that while volume surges, price swings are caused by significant amount of trading activities from various agents.

Beyond volume, researchers have probed how price formation and return patterns reflect the heterogeneous processing of macro news. Han (2025) develops a dynamic noisy rational-expectations model to explain empirical anomalies around announcement days. In the paper, it mentioned that when some investors are better informed than others, informed traders’ forecast revisions only imperfectly trickle into average expectations, allowing “noise” to build up in prices prior to the news. Once the announcement occurs and public information resets beliefs, that accumulated noise gets corrected – resulting in price reversals or unusual return patterns on announcement day. In the later DGP, we will find ways to capture the differences with simulated noises and adjust the scaling parameter on the variance according to various agents.

When a major macroeconomic surprise happens in the market, investors do not all adjust their portfolios in the same way. A growing number of recent research show how heterogeneity in beliefs and constraints leads to different portfolio rebalancing in response to macro news. One important insight is that announcements often synchronize certain trading actions by resolving uncertainty and reducing disagreement. Ying (2021) documents this phenomenon around U.S. monetary policy announcements: using high-frequency equity and options data, he finds that the open interest in index options drops by over 50 $\%$ on FOMC announcement days, implying that a wave of investors are unwinding their pre-positioned trades once the news is publicly available.

Another group of researchers examines how investors move money between asset classes (stocks, bonds, etc.) after macro shocks – and finds that heterogeneity in investor type is crucial. Lu and Wu (2023) focus on institutional investors who allocate across multiple asset classes and propose that their rebalancing activity is a key driver of stock market reactions to interest-rate surprises. The intuition is that certain institutions will sell equities and buy bonds when rates rise unexpectedly, to maintain portfolio targets or exploit higher yields, whereas other investors like equity-focused retail traders may not respond to substituting between asset classes. This heterogeneous behavior will further create an uneven impact: stocks heavily held by multi-asset allocators are more sensitive to the rebalancing sell-off than stocks. However, this is a more complicated set of questions that need to be investigated further and can’t be captured with a simpler DGP setting.

Perhaps the most practical manifestation of heterogeneous responses to macro news is the contrast between retail investors and institutional investors. A number of post-2020 studies have zeroed in on how these trader groups differ in anticipation of and reaction to economic announcements. Evidence from stock markets suggests a clear divergence in behavior before the news breaks. Jia et al. (2023) analyze proprietary Chinese stock exchange data around a high-profile monthly macro announcement. They find that institutional investors actively de-risk ahead of the announcement: on average, institutions cut their overall stock exposure in the days leading up to the news, while selectively tilting into smaller-cap, riskier stocks to seek higher potential upside.

And interestingly, many individuals see an asset’s rally on good news and decide to bet on a reversal, rather than update their fundamental outlook. The authors demonstrate that this behavior is predictable and uninformed: simple trading strategies that take the opposite side of aggregate retail order flow around news events are profitable, indicating that retail trades systematically mis-react to macro news. This also reinforces the hypothesis that different market participants have differential abilities to interpret public information papers and institutional investors typically have better models and expertise to quickly digest an employment report or GDP print, whereas retail traders may be swayed by heuristics, recent price movements, or media sentiment. As historical trades and order data are highly restricted by market makers and exchanges, there are limited amounts of publicly available data so we decided to analyze these hypotheses and claims regarding different investors’ reactions to shocks with simulated data. All the literature provides us a theoretical background for the parameter tuning for the DGP.

We classify traders into four types with fixed risk aversion levels: Retail, Pension Funds, Institutional Investors, and Hedge Funds. Risk appetite and market information levels increase progressively across these types. Hedge funds, with the highest risk tolerance and better information access, trade more aggressively on news, placing larger orders. In contrast, retail trade less frequently and in smaller quantities.

Before scheduled economic announcements, platforms like Bloomberg post publish consensus forecasts. For example, if the market expects the next month’s CPI to be 3.3%. And if the actual CPI is 3.5%, the surprise, defined as the difference between the actual and the expected value, will be $\text{Surprise}=\text{Actual}-\text{Expected}=3.5\%-3.3\%=+0.2\%.$

A good surprise means the data is better than expected. However, in the case of CPI, a higher CPI implies higher interest rates, reduced spending, and an increased cost of borrowing; hence, a positive surprise number is actually considered a bad surprise.

In the DGP, we assume to have two types of asset. Risk Free Asset; Risk On Inflation Sensitivity Asset.

Building on the Capital Asset Pricing Model (CAPM), we aim to measure how asset returns are determined not only by traditional market risk but also by unexpected macroeconomic shocks. In the standard CAPM, the excess return on an asset is explained solely by its sensitivity to the market premium. While, in this project, asset returns also depend on macroeconomic shocks, which decomposed into temporary and permanent components.

Here, $\beta_{i}(R_{m}-R_{f})$ is the traditional market premium while $\beta_{i}$ measures the asset’s sensitivity to overall market movements. $\gamma_{i}S$ captures the asset’s response to the overall CPI surprise. And $\delta_{i}S^{P}$ measures the permanent component of the CPI surprise. And notice there might be structural changes in the market condition after some huge shocks, we then decide to use$\delta_{i}$ to measure how much of the asset’s return is affected by these long lasting changes, ie. huge inflation rate that could alter the entire market. And the $alpha_{i}$ here accounts for asset-specific characteristics not captured by other factors.

In our simulation, each trader selects an order size from a discrete set (e.g., small, medium, large). The instantaneous utility for trader $k$ choosing order size $\ell\in\{\text{small},\,\text{medium},\,\text{large}\}$ is modeled as:

Here we have $c_{0,\ell}$ is a baseline utility term for order size $\ell$; $\mathrm{RA}_{k}$ is trader $k$’s risk aversion level; $\bigl{|}\mathrm{Surprise}\bigr{|}$ is the magnitude of the CPI surprise; $\mathrm{Liquidity}$ is a level of market liquidity.

Given we can measure the utility of each trader’s transaction, we further construct the choice probabilities as a softmax function for each trader.

Here the denominator is the sum of all probabilities, which sum up to 1. And exponentiating the utility magnifies differences, meaning a higher utility leads to a substantially higher probability of the trader $k$ choose to trade size $l$. In parallel to the discrete choice of the order size, we capture how traders allocate their wealth between assets and in which order quantity

In the DGP, we assume that each trader’s risk aversion parameter, $\gamma_{k}$, is constant over multiple time periods. In the graph we assume Retail Trader has the highest risk aversion $\gamma_{k}=3$, Pension has $\gamma_{k}=2$, Institution has $\gamma_{k}=1.5$, and Hedge Funds has $\gamma_{k}=1$. Traders maximize the expected utility of their final wealth, where utility is given by the isoelastic (CRRA) function:

with final wealth defined as:

Since the $W_{0}$ is the initial wealth, we can factor out $W_{0}$ from the utility expression:

Then we substitute into the utility function

Because $W_{0}^{1-\gamma_{k}}/(1-\gamma_{k})$ is a constant factor, solving for the optimal $x$ is equivalent to

In our model, we define $R_{r}$ follows a normal distribution with mean 0.1 and standard deviation 0.2. Because a closed‐form solution for the CRRA utility optimization generally does not exist under this distributional assumption, we thus want to solve this with a numerical (Monte Carlo) approach to approximate the trader’s expected utility and select an optimal allocation.

Monte Carlo Approximation

We implement a discrete grid search over $x$ to approximate the optimal solution. The interval $[0,1]$ is divided into increments of 0.05, generating candidate allocations. For each $x$, we draw multiple i.i.d. samples of the risky return $R_{r}$ from a normal distribution with mean 0.1 and standard deviation 0.2, adjusting the standard deviation based on each trader’s information level (more informed traders have lower noise).

For each draw, we compute the final wealth $W_{\mathrm{final}}$ and evaluate the trader’s CRRA utility $U(W_{\mathrm{final}})$. We then estimate $\max\mathbb{E}[U(W_{\mathrm{final}})]$ by averaging over all samples.

An alternative approach is to optimize the expectation under exponentiation, leveraging the analytical result for normally distributed variables:

However, directly substituting $E[e^{X}]$ into the utility function may misrepresent risk preferences, as expected utility depends on the full return distribution, and after some testing and exploration, we decided to settle with Monte Carlo for the DGP.

One of the primary objectives of this project is to explore how differences in trader behavior and information affect portfolio returns. In our framework, we hypothesize that:

• •

  Institutional traders and hedge funds, with professional quantitative agents have more accurate information and lower risk aversion, are likely to achieve higher returns than other traders.

• •

  Retail traders, who are less informed, tend to earn the lowest among the four types of traders.

• •

  Pension funds are more conservative than institutional traders and hedge funds but are more informed than retail traders, they may choose safer assets and has better portfolio performance than retail.