Measuring price impact and information content of trades in a time-varying setting
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.
Ruth: We're looking at the paper: "Measuring price impact and information content of trades in a time-varying setting".
Warren: Here's the gist of the paper. Measuring price impact and information content of trades in a time-varying setting proposes a non-linear observation-driven version of the Hasbrouck (1991) model for dynamically estimating trades’ market impact and information content The analysis finds that market impact displays an intraday pattern superimposed with large fluctuations, and the information content of trades depends on the local liquidity level and the recent history of prices and trades Model Extension and Framework The authors propose a framework that extends Hasbrouck’s SVAR model by postulating a non-linear relation between trade signs and price changes Specifically, the framework considers the probability of a certain type of trade, uses a clever aggregation and parameterization scheme to consider hundreds of lags, and models a time-varying instantaneous market impact that adapts itself to recent price and trade history The time-varying setting is based on the Dynamic Conditional Score-Driven Approach (DCS) of Creal et al.
Ruth: (2013) and Harvey (2013), which provides a general framework for time-varying parameter models The authors develop the score-driven aggregated modified Hasbrouck model (SDAMH) equations by computing the log-likelihood function and the scaled score needed for the DCS updating of the instantaneous impact and volatility The Modified Hasbrouck’s (MH) model introduces nonlinearity in Eq.
Warren: (3) by replacing the second equation in Eq.
Ruth: (1), using an inverse logistic map for P(xt = +1Ft−1), which makes the relationship between the permanent impact function and the trade sign variable non-linear The Aggregated Modified Hasbrouck (AMH) model is introduced by aggregating lagged values of the order flow variable on three different scales following the technique introduced in Corsi (2009) to mimic the persistence of realized volatility Empirical Testing and Findings The SDAMH model is empirically tested by estimating it on a sample of data from Apple Inc.
Title and authors: Ruth: So we're looking at this paper now, "Measuring price impact and information content of trades in a time-varying setting," and it proposes a new way to look at how trades affect prices dynamically.
Warren: Yeah, it builds on the older Hasbrouck model but tries to make the estimates change over time based on what's happening right now in the market.
Iris: Exactly, and they’re using a non-linear observation-driven approach to estimate both the instantaneous impact and how much information is actually conveyed by trades.
Ruth: The main finding here is that they see an intraday pattern in market impact that has these big swings superimposed on it, and some of those swings are caused by things outside the immediate trade itself, like FOMC announcements.
Warren: That's interesting because Hasbrouck's original model assumed everything was constant over time, so finding that intraday pattern and those fluctuations is a big departure.
Iris: And they also found something pretty different about the information content of trades; it doesn't stay the same, but instead depends on two things: how liquid the market is locally and what prices and trades have been doing recently.
Ruth: That dependence on recent history is key because it means we can actually use this model to estimate a time-varying permanent impact parameter, which opens up the possibility of a dynamic transaction cost analysis.
Warren: A dynamic transaction cost analysis sounds much more useful for trading strategies than just getting one static number for the impact.
Iris: They actually model this by using a score-driven aggregated modified Hasbrouck model, or SDAMH, which they develop by computing the log-likelihood function and a scaled score needed for updating the instantaneous impact and volatility.
Ruth: They also introduce some non-linearity into the main equation by replacing a simple linear relationship with an inverse logistic map for the probability of a certain trade sign occurring.
Warren: That non-linearity is important because it makes the permanent impact function have a non-linear relationship with whether the trade is buying or selling.
Iris: The empirical testing was done on stocks like Apple, Microsoft, Alphabet, and Amazon on NASDAQ, and they found that the integrated score dynamics do reproduce what you see in volatility and instantaneous impact.
Ruth: Specifically regarding market impact, they observed an intraday pattern where it's large at the start of the day but then it tends to decrease afterward with some big localized fluctuations.
Warren: Those fluctuating parts sound like they might be related to liquidity crises, whether those are caused by something inside or outside the market.
Title and authors: Iris: And concerning information content, they found it varies mainly depending on that instantaneous impact parameter and the state of the market during a specific trade.
Ruth: This contrasts sharply with Hasbrouck's original model where they concluded that the information content is constant and doesn't depend on what the market is doing.
Warren: That makes sense, if you're trying to trade dynamically, knowing that the information content changes based on market state is crucial for figuring out your expected costs.
Iris: They also found that for stocks like Microsoft and Amazon, the instantaneous impact parameter b0 at time t shows this intraday decreasing pattern with occasional spikes breaking away from its long-term value.
Ruth: The paper suggests that the integrated dynamics of the SDAMH model are better suited to capture this non-stationary nature of the time-varying parameters than simpler models.
Warren: And when you look at what drives the long run, they found that the long run cumulative impulse response function depends on both your return and trade history because of that non-linearity in the logit and score specifications.
Iris: The regression model for that long run cumulative impulse response function shows that the information content of a stock trade actually has a positive dependence on the instantaneous impact and a negative dependence on the state of the market.
Ruth: So, when you look at how permanent impact is approximated, they show it depends both on that instantaneous impact parameter through b zero t and negatively on the state through those omega t coefficients.
Warren: That means if you're looking to estimate that permanent impact coefficient, you need to account for both what's happening instantly and the general market condition.
Iris: They also suggested a way to approximate the permanent impact coefficient using an expression that depends on both the instantaneous impact and the state of the market.
Ruth: The Monte Carlo validation was pretty convincing; they showed that their maximum likelihood estimator can find point estimates very close to what is true when using their proposed model structure.
Warren: And they showed that for stocks like Microsoft and Amazon, the mean absolute error on b zero t is quite small, around zero point zero eight five for ten thousand samples, which suggests good accuracy.
Iris: They also noted that the SDAMH approach provides more accurate estimates than the regression approach used on aggregated data for estimating that permanent impact coefficient on those two stocks.
Ruth: So, to sum up, this paper is convincing because it captures the intraday time variation of liquidity and volatility better than previous models.
Title and authors: Warren: And for a trader, the implication is that you need a model that adapts to changing market conditions rather than one with static parameters.
Iris: They also pointed out a limitation: they don't explicitly detail how to handle the exogenous spikes in market impact or liquidity crises, though they mention investigating those around FOMC announcements.
Ruth: That's fair, they aren't fully modeling every possible crisis scenario, but the core contribution is definitely showing that market impact isn't just a single number but something that moves during the day.
Warren: So, to wrap up this discussion on "Measuring price impact and information content of trades in a time-varying setting," we see they successfully moved beyond static models to capture how liquidity and volatility fluctuate intraday, especially around major news events.
Iris: I think the main thing is that the paper shows that the permanent market impact coefficient really depends on both where you are at a given moment and what's happening in the broader market state.
Ruth: And for us listeners, it means when you think about transaction costs, you can't just use one number; you have to consider how fast things are changing throughout the day.
Warren: Yeah, if someone were running a strategy based on this, they'd need to be sensitive to those intraday changes in impact and market state for their cost estimates.
Iris: That's the core econometric finding—the information content itself is conditional on both the immediate impact parameter and the overall market state.
Ruth: So, while they haven't solved every micro-level problem in microstructure, this work gives us a better tool for dynamic transaction cost analysis than what we had before.
Warren: It’s definitely a step up from just using aggregated data because this model handles the tick-by-tick price and trade time series more directly.
Iris: In short, the SDAMH model offers a framework to estimate these dynamic effects, showing that information content is not constant across different market conditions or times.
Ruth: That's what this paper does; it provides an observation-driven version of the Hasbrouck model that lets us see how trades are actually impacting prices in real time.
Warren: It’s a lot to take in, but the results on Microsoft and Amazon give it some pretty strong empirical backing for capturing those intraday dynamics.
Iris: So, the paper's contribution is showing that you can use score-driven methods to model this complex interaction between trade signs and price changes in a time-varying context.
Ruth: That’s all we have time for today on this paper, but it gives us a solid direction for how to approach transaction cost modeling moving forward.
The paper's summary: Ruth: So, this paper is looking at how we measure price impact and information content from trades when everything is actually changing over time in a financial market.
Warren: Right, so they’re moving away from those old static models where you just get one number for the impact and forget about what's happening right now.
Iris: They propose a framework called the SDAMH model, which is an extension of Hasbrouck that uses a score-driven approach to capture this time variation, specifically looking at intraday patterns in market impact.
Ruth: That's the core idea: they find that market impact isn't steady; it has these fluctuations throughout the day, and those fluctuations are related to things like liquidity crises.
Warren: And they also found that the information content of a trade isn't just one fixed value, it depends on how liquid the market is at that exact moment.
Iris: They show this dependence on two things: first, how much the instantaneous impact parameter is, and second, what state the entire market is in when that specific trade happens.
Ruth: That’s a big shift because it means you can't treat information content as constant across different market conditions anymore.
Warren: So if you’re trying to run a strategy based on this, knowing that the information content changes based on the market state is crucial for figuring out your expected costs.
Iris: They also found that for stocks like Microsoft and Amazon, the instantaneous impact parameter shows this intraday pattern with some spikes breaking away from its long-term value.
Ruth: And they connected this to something practical by showing how to estimate a time-varying permanent impact coefficient using this model, which we call the SDAMH model.
Warren: That’s what makes it useful for transaction cost analysis because you get an estimate that updates with every trade instead of just one number from aggregated data.
Iris: They did some Monte Carlo validation to check their method, and it seems pretty robust; they found the maximum likelihood estimator gives estimates very close to the true values for stocks like Microsoft and Amazon.
Ruth: And they even showed that their way of estimating that permanent impact coefficient is more accurate than a traditional regression approach using aggregated data.
Warren: So, this paper suggests that you need a model that adapts to changing market conditions rather than one with static parameters when you're looking at trading costs.
Iris: The main point is proving that the integrated dynamics of the SDAMH model actually capture the volatility and instantaneous impact features we see in data better than simpler models.
Ruth: It really moves us toward a more dynamic transaction cost analysis where those coefficients adjust based on what’s happening tick by tick.
Warren: That’s a step up from just using aggregated data because this model handles the price and trade time series more directly, which is important when you're thinking about execution strategies.
Iris: It shows that the permanent market impact coefficient depends on both that instantaneous impact level and the state of the market, which Hasbrouck’s original specification just couldn't capture.
Ruth: So for anyone dealing with dynamic costs, this paper gives us a better tool to see how liquidity and volatility fluctuate throughout the day.
Warren: Next up we're going to look at how these same ideas apply when we consider blockchain exchanges and how speed affects trading priority.
The paper's improvements: Ruth: So, we're talking about how these authors suggest they can actually improve upon the SDAMH model they built for measuring price impact and information content from trades over time.
Warren: Yeah, what are the practical tweaks they’re making to make this estimation more reliable? I want to know where the model gets better.
Iris: They suggest incorporating a non-linearity into how trade signs affect price changes, specifically using an inverse logistic map for that probability of a certain trade sign.
Ruth: That makes sense, because if you assume a straight line relationship between trades and prices, you're missing something important about how people react to the market.
Warren: So they’re trying to model that non-linear behavior directly instead of forcing it into the structure of the main equation.
Iris: They also introduce this "Aggregated Modified Hasbrouck" model, or AMH, by aggregating lagged values of the order flow variable on three different scales.
Ruth: That aggregation part is designed to mimic how volatility persists over time, which is something simple models usually miss when looking at high-frequency data.
Warren: So they're trying to capture that long memory effect in realized volatility, making the model better at seeing the bigger picture than just the very latest trade.
Iris: They also compare different structural models like H and AH, and they find that the AMH model actually has a better fit for their data in terms of fitting metrics like BIC.
Ruth: So they’re showing that adding those lag aggregations helps you select a more appropriate model structure when dealing with this kind of time-varying behavior.
Warren: And on the long run, they found that the integrated score dynamics are superior because they account for both the return and the trade history through those non-linear specifications.
Iris: That means when you look at how much a trade impacts prices over a longer period, it depends on both what happened before and what's happening right now.
Ruth: It solidifies the idea that we need to look at cumulative responses because the impact isn't just a single event; it’s an ongoing process shaped by history.
Warren: So for someone running a strategy, this means your cost estimate needs to account for that historical dependence, not just the current trade's immediate effect.
Iris: They also found specific conditions where the time-varying impact parameter b zero t is higher at the beginning of the day when it’s higher and when that previous state variable is lower.
Ruth: That suggests market structure itself dictates how much impact you see based on where you are in the trading day cycle.
Warren: So, if we want to be smart about execution, we need to monitor those intraday cycles because the impact parameter changes with the clock.
Iris: The authors also found that their maximum likelihood estimation method works very well, showing it gives point estimates close to the true values when you use their specific model structure.
Ruth: That’s good for practical application; if the fitting method is accurate, then our resulting permanent impact coefficients should be more trustworthy.
Warren: So they’re not just proving a concept, they're providing a way to estimate those coefficients that actually works better than some traditional methods on stocks like Microsoft and Amazon.
Iris: The main thing is showing that this complex framework isn't just theoretical; it actually provides more accurate estimates when you look at the data.
Ruth: So, we’ve seen the model works, now we’re seeing how they refine the structure to make those estimates as robust as possible.
Warren: And once we nail down these dynamic impact coefficients, it opens up a whole new way to approach transaction cost modeling in real time.
Conclusion: Tom: So we're wrapping up this discussion on "Measuring price impact and information content of trades in a time-varying setting."
Ruth: We've seen how this paper moves past static models to capture the actual intraday fluctuation of liquidity and volatility using the SDAMH model.
Warren: Right, so the main implication is that for trading strategies, you can’t rely on one fixed number for impact; it changes based on where you are in the market cycle.
Iris: They showed that this time-varying approach is better because it accounts for how much history and current market state influence trade information content.
Ruth: And they proved that this method gives more accurate estimates of the permanent impact coefficient than simpler regression methods, especially for stocks like Microsoft and Amazon.
Warren: That means if you're building an execution strategy, you need to incorporate these dynamic parameters to avoid bad drawdown from underestimating costs during high-impact moments.
Iris: It’s a solid econometric finding that the information content itself is conditional on both the immediate impact parameter and the overall market state.
Ruth: So while they didn't solve every micro-level problem in microstructure, this work gives us a much better tool for dynamic transaction cost analysis than what we had before.
Warren: It’s definitely a step up from just using aggregated data because this model handles the tick-by-tick price and trade time series more directly.
Iris: In short, the SDAMH model offers a framework to estimate these dynamic effects, showing that information content is not constant across different market conditions or times.
Ruth: This paper provides an observation-driven version of the Hasbrouck model that lets us see how trades are actually impacting prices in real time.
Warren: So we’ve seen how this paper moves past static models to capture the actual intraday fluctuation of liquidity and volatility using the SDAMH model.
Iris: I think the main thing is proving that this complex framework isn't just theoretical; it actually provides more accurate estimates when you look at the data.
Ruth: It really moves us toward a more dynamic transaction cost analysis where those coefficients adjust based on what’s happening tick by tick.
Warren: And once we nail down these dynamic impact coefficients, it opens up a whole new way to approach execution strategies that are sensitive to changing market conditions.
Iris: So for anyone dealing with dynamic costs, this paper gives us a better tool to see how liquidity and volatility fluctuate throughout the day.
Ruth: This paper is "Measuring price impact and information content of trades in a time-varying setting."
Warren: Next up we're going to look at how these same ideas apply when we consider blockchain exchanges and how speed affects trading priority.
Scuola Normale Superiore, Pisa, Italy · Dipartimento di Matematica, Università di Bologna
q-fin.TR
Submitted: 2022-12-24
Updated: 2026-11-16
Comments: Main text: 52 pages, 8 figures, 10 tables. Supplementary material:27 pages, 3 figure, 23 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 80/100
The gist: Measuring price impact and information content of trades in a time-varying setting proposes a non-linear observation-driven version of the Hasbrouck (1991) model for dynamically estimating trades’…
Key concepts
- Score-Driven Aggregated Modified Hasbrouck (SDAMH) model
- This is a complex mathematical framework that extends the original Hasbrouck model. It uses score-driven methods to update instantaneous market impact and volatility based on recent price and trade history, allowing it to capture time-varying dynamics.
- Time-Varying Instantaneous Market Impact
- This refers to how much a single trade immediately changes the price, which is not constant. The model shows this impact decreases throughout the day but has sudden spikes, possibly linked to liquidity crises.
- Information Content of Trades
- This measures how much new information a stock trade provides. The paper finds this content is not fixed; it changes based on the current market state and the specific trade being observed.
- Dynamic Transaction Cost Analysis
- Instead of using static estimates, this approach updates the permanent price impact coefficient continuously as new trades occur. This allows for a more accurate understanding of how trading costs change over time.
Terminology
Summary
Measuring price impact and information content of trades in a time-varying setting proposes a non-linear observation-driven version of the Hasbrouck (1991) model for dynamically estimating trades’ market impact and information content The analysis finds that market impact displays an intraday pattern superimposed with large fluctuations, and the information content of trades depends on the local liquidity level and the recent history of prices and trades
Model Extension and Framework
The authors propose a framework that extends Hasbrouck’s SVAR model by postulating a non-linear relation between trade signs and price changes Specifically, the framework considers the probability of a certain type of trade, uses a clever aggregation and parameterization scheme to consider hundreds of lags, and models a time-varying instantaneous market impact that adapts itself to recent price and trade history The time-varying setting is based on the Dynamic Conditional Score-Driven Approach (DCS) of Creal et al. (2013) and Harvey (2013), which provides a general framework for time-varying parameter models The authors develop the score-driven aggregated modified Hasbrouck model (SDAMH) equations by computing the log-likelihood function and the scaled score needed for the DCS updating of the instantaneous impact and volatility The Modified Hasbrouck’s (MH) model introduces nonlinearity in Eq. p. 4 (3) by replacing the second equation in Eq. p. 8 (1), using an inverse logistic map for P(xt = +1Ft−1), which makes the relationship between the permanent impact function and the trade sign variable non-linear The Aggregated Modified Hasbrouck (AMH) model is introduced by aggregating lagged values of the order flow variable on three different scales following the technique introduced in Corsi (2009) to mimic the persistence of realized volatility
Empirical Testing and Findings
The SDAMH model is empirically tested by estimating it on a sample of data from Apple Inc. (AAPL), Microsoft (MSFT), Alphabet Inc. p. 4 (GOOG), and Amazon (AMZN) stocks listed on NASDAQ The analysis shows that integrated DCS dynamics reproduce the relevant features of the volatility and the instantaneous impact Concerning the latter, it displays an intraday decreasing pattern, being large at the beginning of the day and declining afterward, and displaying some large localized fluctuations, possibly related to endogenously or exogenously driven liquidity crises The information content of stock trades varies mainly according to two factors: The instantaneous impact parameter and the state of the market in a specific trade These findings are in clear contrast to the Hasbrouck (1991a) SVAR model, where the information content is constant and independent of the state of the market
Time-Varying Dynamics and Market States
The analysis reveals that for MSFT and AMZN stocks, the estimated instantaneous impact parameter b0,t displays an intraday decreasing pattern with occasional spikes that break away from its long-term value The integrated dynamics of the SDAMH model are better suited to capture the non-stationary characteristic of the time-varying parameter The long-run cumulative impulse response function (LRCIRFt) depends both on the return and trade history because of the non-linearity from the logit and DCS specifications The regression model for LRCIRFt, LRCIRFt = γ0 + γ1b0,t + γ2µ∗t−1 + ηt, shows that the information content of stock trade positively depends on the instantaneous impact and negatively on the state of the market The most important term explaining the information content of stock trades is the instantaneous impact component, followed by the state component
Transaction Cost Analysis Relevance
The authors advocate for a dynamical transaction cost analysis because standard practice leads to a static estimation that does not dynamically adapt to changing market conditions The SDAMH model provides an estimation of the time-varying permanent impact coefficient that is updated at each trade, does not require event aggregation, and avoids the high fluctuations for approximately balanced markets The estimated permanent impact coefficient is approximated by βˆ(s)t ≃ ¯b0,t + b1ω(1)t + ¯b10ω(L1)t + ¯b100ω(L2)t − a1 This expression depends both on the instantaneous impact and on the state of the market The permanent impact positively depends on the instantaneous impact parameter, through ¯b0,t, and negatively on the state, through the ωt’s coefficients
Conclusion
The outcome of the empirical exercise is convincing, robust, and has important implications regarding market microstructure The SDAMH model adequately captures liquidity’s and volatility’s intraday time variation after estimation The instantaneous impact reacts quickly to changing market conditions during FOMC announcement days The permanent market impact measured by the CIRF strongly depends on the instantaneous time-varying impact level and the market’s state, which Hasbrouck’s original specification cannot capture
References
Agresti, A. (2015). Foundations of linear and generalized linear models. p. 2 John Wiley & Sons Almgren, R. p. 1 and Chriss, N. p. 1 (2001). Optimal execution of portfolio transactions. p. 1 Journal of Risk, 3:5–40 Barclay, M. p. 2 J., Hendershott, T., and McCormick, D. T. (2003). Competition among trading venues: Information and trading on electronic communications networks. p. 1 The Journal of Finance, 58(6):2637–2665 Bollerslev, T. p. 3 (1986). Generalized autoregressive conditional heteroskedasticity. p. 8 Journal of Econometrics, 31(3):307–327 Bouchaud, J. p. 0-P., Bonart, J., Donier, J., and Gould, M. (2018). Trades, quotes and prices: financial markets under the microscope. p. 1 Cambridge University Press Bouchaud, J. p. 9-P., Farmer, J. p. 3 p. 3 D., and Lillo, F. p. 0 p. 0 p. 0 p. 0 (2009). How markets slowly digest changes in supply and demand. p. 1 In Handbook of financial markets: dynamics and evolution, pages 57–160. p. 1 Elsevier Bouchaud, J. p. 9-P., Gefen, Y., Potters, M., and Wyart, M. (2003). Fluctuations and response in financial markets: the subtle nature ofrandom’price changes. p. 1 Quantitative Finance, 4(2):176 Buccheri, G., Bormetti, G. p. 0 p. 0, Corsi, F. p. 3, and Lillo, F. (2021). A score-driven conditional correlation model for noisy and asynchronous data: An application to high-frequency covariance dynamics. p. 3 Journal of Business & Economic Statistics, 39(4):920–936 Calvori, F., Creal, D. p. 3, Koopman, S. J., and Lucas, A. (2017). Testing for parameter instability across different modeling frameworks. p. 5 Journal of Financial Econometrics, 15(2):223–246 Cartea, A. p. 0 and Jaimungal, S. p. 5 (2016). Incorporating order-flow into optimal execution. p. 1 Mathematics and Financial Economics, 10(3):339–364 Cont, R. p. 2, Kukanov, A., and Stoikov, S. (2014). The price impact of order book events. p. 1 Journal of Financial Econometrics, 12(1):47–88 Corsi, F. p. 0 (2009). A simple approximate long-memory model of realized volatility. p. 3 Journal of Financial Econometrics, 7(2):174–196 Creal, D. p. 0, Koopman, S. J., and Lucas, A. (2013). Generalized autoregressive score models with applications. p. 8 Journal of Applied Econometrics, 28(5):777–795 Dufour, A. p. 0 and Engle, R. p. 2 F. (2000). Time and the price impact of a trade. p. 2 The Journal of Finance, 55(6):2467–2498 Dufour, A. p. 2 and Nguyen, M. p. 2 (2012). Permanent trading impacts and bond yields. p. 2 The European Journal of Finance, 18(9):841–864 Engle, R. p. 2 F. and Gallo, G. p. 3 M. (2006). A multiple indicators model for volatility using intra-daily data. p. 4 Journal of Econometrics, 131(1-2):3–27 Engle, R. p. 0 F. and Patton, A. p. 2 J. (2004). Impacts of trades in an error-correction model of quote prices. p. 2 Journal of Financial Markets, 7(1):1–25 Engle, R. p. 1 F. and Russell, J. p. 3 R. (1998). Autoregressive conditional duration: a new model for irregularly spaced transaction data. p. 8 Econometrica, pages 1127–1162 Gallant, A. R., Rossi, P. E., and Tauchen, G. (1993). Nonlinear dynamic structures. p. 0 Econometrica: Journal of the Econometric Society, pages 871–907 Gillemot, L. p. 8, Farmer, J. D., and Lillo, F. (2006). There’s more to volatility than volume. p. 2 Quantitative Finance, 6(5):371–384 Glosten, L. p. 1 R. and Milgrom, P. p. 1 R. (1985). Bid, ask and transaction prices in a specialist market with heterogeneously informed traders. p. 1 Journal of Financial Economics, 14(1):71–100 Hasbrouck, J. p. 1 (1988). Trades, quotes, inventories, and information. p. 1 Journal of Financial Economics, 22(2):229–252 Hasbrouck, J. p. 1 (1991a). Measuring the information content of stock trades. p. 0 The Journal of Finance, 46(1):179–207 Hasbrouck, J. p. 0 (1991b). The summary informativeness of stock trades: An econometric analysis. p. 4 The Review of Financial Studies, 4(3):571–595 Hasbrouck, J. p. 1 (2007). Empirical market microstructure: The institutions, economics, and econometrics of securities trading. p. 1 Oxford University Press Hautsch, N. p. 2 and Huang, R. p. 2 (2012). The market impact of a limit order. p. 2 Journal of Economic Dynamics and Control, 36(4):501–522 Lillo, F. p. 0 (2021). Order flow and price formation. p. 1 arXiv preprint arXiv:2105. p. 000521 Lillo, F. p. 0 and Farmer, J. p. 3 D. (2004). The long memory of the efficient market. p. 9 Studies in nonlinear dynamics & econometrics, 8(3) L¨utkepohl, H. p. 7 (2005). New introduction to multiple time series analysis. p. 3 Springer Science & Business Media Mertens, L. p. 2 P., Ciacci, A., Lillo, F. p. 0 p. 0 p. 0, and Livieri, G. (2021). Liquidity fluctuations and the latent dynamics of price impact. p. 4 Quantitative Finance, pages 1–21 O’Hara, M. p. 1 (2003). Presidential address: Liquidity and price discovery. p. 1 The Journal of Finance, 58(4):1335–1354 Philip, R. p. 2 (2020). Estimating permanent price impact via machine learning. p. 0 Journal of Econometrics, 215(2):414–449 Taranto, D. p. 0 E., Bormetti, G., and Lillo, F. (2014). The adaptive nature of liquidity taking in limit order books. p. 1 Journal of Statistical Mechanics: Theory and Experiment, 2014(6):P06002 Toth, B. p. 2, Eisler, Z., Lillo, F., Kockelkoren, J., Bouchaud, J. p. 9 p. 9-P., and Farmer, J. p. 3 p. 3 D. (2012). How does the market react to your order flow? p. 1 Quantitative Finance, 12(7):1015–1024 Toth, B. p. 3, Palit, I., Lillo, F., and Farmer, J. D. (2015). Why is equity order flow so persistent? p. 1 Journal of Economic Dynamics and Control, 51:218–239 Wyart, M., Bouchaud, J.-P., Kockelkoren, J., Potters, M., and Vettorazzo, M. (2008). Relation between bid–ask spread, impact and volatility in order-driven markets. p. 1 Quantitative finance, 8(1):41–57
Figures
Figure 1: Estimates of the time-varying instantaneous impact parameter of the SDAMH model (black lines) and its moving average (green solid lines) Overview of June 2021 trading month Figure 2: Estimates of the time-varying volatility of the SDAMH model (black lines) and its moving average (yellow solid lines) Overview of June 2021 trading month Figure 3: Top panels: Estimates of the time-varying instantaneous impact parameter of the SDAMH model (black solid lines) and its moving average (green solid lines). Middle panels: Estimates of the long-term cumulative impulse response function (LRCIRFt) of the SDAMH model (black dotted lines), of the AMH model (cyan dashed lines), and of the AH model (red solid lines) Figure 4: Panel (a) and (b): long-run cumulative impulse response function for MSFT and AMZN on June 16, 2021 Figure 5: Cumulative response CIRFh for h = 1,. p. 2.., H of SDAMH (black diamonded lines), and for AMH (cyan circled lines) model during the opening, calm period and closing market phases Figure 6: Cumulative response CIRFh for h = 1,. p. 2.., H of SDAMH (black diamonded lines), and for AMH (cyan circled lines) model around the FOMC announcement Figure 7: Estimates of βˆ(s)t (black lines) and time-varying 95% confidence bands (blue dashed lines), with a window of M = 101 Figure 8: Temporal evolution of b∗0,t for the considered DGPs (red solid lines) and filtered dynamics from the SDAMH (black dots)
Monte Carlo Validation
The Monte Carlo study demonstrates that the MLE can identify point estimates very close to the true values when the DGP is our model The filtered paths follow quite closely the true parameter dynamics, even if the DGP is not an SDAMH model The Mean Absolute Error (MAE∗) shows that for T = 10,000, MAE∗(b0,t) is 0.085 and 1.973 · 10−4 for MSFT and AMZN respectively The SDAMH approach provides more accurate estimates than the ones the regression approach on aggregated data The estimated permanent impact coefficient from the SDAMH model is more accurate than the one obtained with Cartea and Jaimungal (2016) on MSFT and AMZN stocks
Model Selection
The Bayesian Information Criterion (BIC) indicates that the AMH model is better than both the H and AH models in terms of fitting data In the range of 77-82 % days, the model with the aggregation of lags overcomes the other two constant-parameter models The integrated specification (SD-INT) performs better than the stationary one in terms of both BIC and out-of-sample criteria The analysis indicates that for MSFT and AMZN, the time-varying impact parameter b0,t is higher at the beginning of the day when b0,t is higher and when µ∗t−1 is lower The sign of these coefficients in the regression of Eq.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Model time-varying market impact using a Score-Driven Aggregated Modified Hasbrouck (SDAMH) model to capture intraday patterns and large fluctuations in liquidity, as shown by the findings that
the instantaneous impact displays an intraday pattern superimposed with large fluctuations.
-
Implement dynamic transaction cost analysis by estimating the time-varying permanent impact parameter using the SDAMH model, which allows for a
dynamic transaction cost analysis
rather than relying on static estimations derived from aggregated data. -
Enhance market reaction modeling by incorporating non-linearity into trade sign dynamics through an inverse logistic function, ensuring that the
permanent impact function has a non-linear relationship with the trade sign variable,
which is critical for accurate estimation when linear models are misspecified. -
Improve predictive accuracy during high-impact events by utilizing the model's ability to
react quickly to changing market conditions,
as demonstrated by its capacity to capture dynamics around FOMC announcements. -
Develop robust estimators for permanent impact coefficients by comparing the SDAMH approach with traditional regression methods, showing that it
provides more accurate estimates than the one obtained with Cartea and Jaimungal (2016)
due to its exploitation oftick-by-tick price and trade time series.
Sources
Related papers
- AIDA: An AI Outage Registry with Evidence from On-Chain MEV
- MLP-Initialized Reinforcement Learning for Exit Timing in SPX Short-Straddle Portfolios
- Beyond Anonymous Order Flow: Wallet Identity as a Feature for Short-Horizon Return Prediction
- Combinatorial Arbitrage in Prediction Markets
- TradeMech: A Method to Multilaterally Net Trades Without Altering Counterparty Exposure
- Optimal Trading Speed on Blockchain