Market-Based Probability of Stock Returns

summary

The gist

We derive the dependence of the market-based mean and variance of returns on the means, variances, and covariance of values and volumes of consecutive trades with shares at an exchange…

In short

The paper derives market-based mean and variance of stock returns based on trade values and volumes, moving beyond simple frequency-based definitions that assume constant trading volume. It shows that return correlation with volume is zero but suggests using correlations with squared volumes for statistical dependence. This approach is crucial for more accurate forecasting in complex markets.

Key concepts

Gross Returns Rg(ti,)
These returns are calculated by dividing the payoff at time ti by the past price at time ti-tau. They are influenced by two random factors: trade prices and dividends, with dividend growth forecasts adding a second layer of randomness to the return calculation.
Market-Based Variance (t)
This variance measures price uncertainty based on the coefficient of variation of trade values (C) and volumes (U), along with their covariance. This contrasts with conventional methods that often ignore volume fluctuations, providing a more realistic measure of risk.
Frequency-Based Approximations
Conventional methods use frequency-based statistics, which assume that the volume of consecutive trades remains constant over time. The paper demonstrates that using these approximations leads to inaccurate forecasts when trade volumes are highly irregular.

Terminology used across episodes

This episode discusses

The paper

Market-Based Probability of Stock Returns · Read on arXiv

Victor Olkhov

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Market-Based Probability of Stock Returns".

Warren: Here's the gist of the paper. We derive the dependence of the market-based mean and variance of returns on the means, variances, and covariance of values and volumes of consecutive trades with shares at an exchange <ref:2302.07935#pg2> The conventional frequency-based definitions of means and variances describe only a limited case when all volumes of consecutive trades are assumed constant <ref:2302.07935#pg5> To forecast market-based variance of returns at horizon T, one should predict the means, variances, and covariance of values and volumes of trades at the same horizon <ref:2302.07935#pg2> We show that the correlation of returns and trade volumes is always zero <ref:2302.07935#pg2> To estimate their statistical dependence, one should empirically calculate the correlation between returns and squares of trade volumes <ref:2302.07935#pg2> Market-Based Return Definition The paper introduces gross returns Rg(ti,) that are determined by the ratio of the payoff x(ti) at time ti to price p(ti- in the past at time ti- <ref:2302.07935#pg4> The payoffs x(ti) are determined by two different random economic factors: by the trade prices p(ti) and dividends D(ti) <ref:2302.07935#pg5> The uncertain forecasts of the growth of the market value of the issuer that predict the payoffs of dividends D(ti) determine the second source of returns RD(ti, in (1.2) randomness <ref:2302.07935#pg5> The paper studies market trade origin of random returns R(ti, in (1.2) and presents the market-based description of their means and variances <ref:2302.07935#pg5> Market-Based Variance Derivation The iterative relations (3.1) determine the market-based n-th statistical moment of price p(t;n)=Em

pn(ti): through the previous statistical moments p(t;m), m=1,2,..

Ruth: n-1 <ref:2302.07935#pg6> For n=2, the relations (3.1-3.3) define market-based variance (t) (3.4) of prices and the 2nd statistical moment p(t;2) (3.4) <ref:2302.07935#pg6> The market-based variance (t) (3.4) of price depends on the coefficient of variation (C) of the trade values C(ti), on the coefficient of variation (U) of the trade volumes U(ti), and on their covariance (C,U) (3.5> The market-based variance (t) (3.4) determines the variance (t,t0) (3.8; 3.9> Limitations of Conventional Approximations The paper shows that for a limited approximation, which assumes that all volumes of consecutive trades are constant, the statistical moments of prices and returns take the usual frequency-based expressions <ref:2302.07935#pg7> The use of frequency-based statistical moments of prices (t;n) (4.2) or returns (t,t0;n) (4.3) implicitly assumes the approximation of constant trade volumes <ref:2302.07935#pg7> The implicit use of a limited approximation of constant trade volumes and relations (t;n) (4.2) or (t,t0;n) (4.3) may result in a low accuracy and wrong forecasts of the corresponding economic and financial models <ref:2302.07935#pg7> Implications for Modeling The lack of econometric observations and assessments of covariances of values and volumes of all transactions made in the economy reduces the number of economic-based predictable statistical moments by the first two <ref:2302.07935#pg8> This limits the accuracy of any forecasts by the accuracy of Gaussian approximations <ref:2302.07935#pg9> To improve reliability of their forecasts and adjust large macroeconomic models like BlackRock's Aladdin, JP Morgan, and the U.S.

Warren: Fed with reality determined by the impact of random trade volumes the developers may use the market-based variances <ref:2302.07935#pg9> The nearest, but still far goal is the economic-based description of the 1st and 2nd order variables <ref:2302.07935#pg9> Correlation Analysis The market-based correlation during the averaging interval Δ (1.3) between the time series of volumes U(ti) of consecutive trades and returns R(ti,t0) with respect to price p(t0) in the past is determined by the correlation of trade volumes U(ti) and prices p(ti) and should obey <ref:2302.07935#pg10> The use of “conventional” expression of the correlation covemp

R(ti,t0), U(ti): (6.7) between returns R(ti,t0) (2.2) and trade volumes U(ti) is based implicitly on the use of the mean return (t,t0;1) (4.3; 6.9> However, if all trade volumes are constant U(ti)=U, then any correlations with time series of constant trade volumes are zero <ref:2302.07935#pg10> The paper proposes the rules for the derivation of higher market-based statistical moments of returns but doesn’t provide the pure prove <ref:2302.07935#pg9> It shows that market-based mean and variance of returns take their usual frequency-based forms if all volumes are assumed constant during averaging <ref:2302.07935#pg7> The returns related to highly irregular trade volumes should be described by the market-based variance <ref:2302.07935#pg9> The empirical studies of statistical dependence between returns and trade volumes or returns and past values should consider market-based correlations of returns and squares of trade volumes or returns and squares of past values <ref:2302.07935#pg9> To improve reliability of their forecasts and adjust large macroeconomic models like BlackRock's Aladdin, JP Morgan, and the U.S.

Ruth: Fed with reality determined by the impact of random trade volumes the developers may use the market-based variances <ref:2302.07935#pg9> References Amaral, L.I., Plerou, V., Gopikrishnan, P., Meyer, M.

Paper summary: Warren: So we've covered how this paper, "Market-Based Probability of Stock Returns," tries to move beyond simple frequency-based returns by looking at the dependence on trade values and volumes, but it also laid out some hard limits on what this method can achieve.

Iris: They showed that if you stick to assuming constant volumes, you just get the standard frequency expressions for means and variances, which isn't very useful when markets are irregular.

Ruth: And they clearly state that the returns related to those highly irregular trade volumes need to be described using this market-based variance approach instead of the old methods if you want accuracy.

Warren: The empirical studies on dependence between returns and volumes should look at correlations with squares of trade volumes or squares of past values, not just returns and raw volumes alone.

Iris: And remember, the paper notes that because there aren't enough econometric observations about all transaction covariances in the whole economy, this limits how accurately we can make economic-based predictions for variances and covariances at any reasonable time horizon.

Ruth: So to wrap up, "Market-Based Probability of Stock Returns" proposes a set of rules for deriving these higher market-based statistical moments, showing they take their usual frequency forms only if all volumes are constant.

Warren: The main point is that returns with highly irregular trade volumes should be described by the market-based variance. And empirical studies need to consider those correlations with squares of trade volumes or squares of past values for a proper look at statistical dependence.

Iris: This paper suggests that to improve the reliability of big models, like those used by BlackRock or JP Morgan, developers can use these market-based variances because they account for the impact of random trade volumes on reality.

Conclusion: Ruth: So we’ve seen how this paper looks at how trade volumes affect return prediction, but now we gotta talk about what this whole thing actually means for the folks who are building big financial models.

Warren: Yeah, it’s about these market-based mean and variance derivations, and it shows that if you just assume constant trade volumes, you end up with the same old frequency-based results.

Iris: Exactly. The paper is making a point about how those standard frequency formulas only work under a very specific condition—when all the trading volumes are steady.

Ruth: That condition is the whole problem, right? If we're dealing with real markets where volume jumps around constantly, those old models just give us inaccurate forecasts for returns.

Warren: And they say that for those irregular trades, you need to use this new market-based variance stuff instead of the conventional way. It’s about capturing that real-world noise in the trading activity.

Iris: But there’s a catch; they admit that because they don't have all the data on every single transaction's correlation, this limits how accurate any economic prediction can be for variances over a long time horizon T.

Ruth: So it’s a tool for better forecasting when things get messy, but it doesn't solve the whole uncertainty problem in finance. It gives us a way to adjust those big models like the ones BlackRock or the Fed use.

Warren: Right, and they point out that if you want to see how this actually works empirically, you should look at correlations between returns and squares of trade volumes, not just raw volumes.

Iris: So the authors are essentially proposing a way to improve those complex macroeconomic models by incorporating the actual randomness of market trading activity into the variance estimates.

Ruth: It’s a step toward making those big financial predictions more robust when you account for how much money is actually moving in and out at any given moment.

Iris: And that leads us next to where this approach stops being just a theoretical exercise and starts becoming a practical tool for risk management.

More episodes

← Home