No conflict, no interest: on the economics of conflicts of interest faced by analysts

Abstract

This paper outlines evolution of the policy response to conflicts of interest analysts face in offering investment advice to investors when the company they follow may also buy merchant banking services from their employer. Both in the US and the UK on a both statutory and common law basis the response has been one of to disclose and let market participants price the implied conflict or simply rebut the advice given. An efficient market can price conflicts and by implication unravel any potential damage to shareholder wealth induced by analysts’ conflicts of interests in this view. I consider the impact the presence of “noise traders” in financial markets may have on the welfare implications of this sort of policy stance. The presence of noise traders casts doubt on the benign impact of conflicts of interest in financial markets. In particular the presence of noise induced variance in analyst’s forecasts implies disclosure based remedies may be ineffective in mitigating the harm of analyst’s conflicts of interest.

Appendix: The IJ model

1.1 Introducing noisy valuations

Much of the subsequent analysis of noise in later sections of the paper is terms of the impact of “noise” on the informational value of the signal concerning value in analysts’ forecasts/recommendations as measured by the AIR. Note price change is decreasing in the AIR because $\sigma_{\varepsilon }^{2}$ is in the denominator of the round bracketed inverse term so that the overall term for price change in square brackets fall as $\sigma_{\varepsilon }^{2}$ rises. As such the AIR is the Bayesian precision of the revision, with respect to earnings information, $\frac{1}{{\sigma_{\varepsilon }^{2} }}$, multiplied by the amount of “noise” in valuation from non-earnings sources. Here $\frac{1}{{\sigma_{\varepsilon }^{2} }}$ is the Bayesian “precision” of the signal about value implied from the analysts’ forecast revision. Now the role of $\sigma_{\varphi }^{2}$ is simply to mask the valuation signal conveyed by analysts forecasts/recommendations. Increases in$\sigma_{\varphi }^{2}$ simply increase the size of the inverse term in rounded brackets and so reduce the size of the overall square-bracketed term for the implied price change.

Reductions in the AIR diminish the power of analysts’ forecast revisions to induce movements in price. To see this consider the case of a perfectly informative “pure” signal about earnings where $\sigma_{\varphi }^{2}$ = 0; that is all “noise” has been purged from the signal regarding value analysts’ forecasts convey. For this case

$$ \Updelta {\text{P}}_{\text{t}} = ({\text{P}}_{\text{t}}^{\text{new}} |\theta_{\text{t}} ) - {\text{P}}_{\text{t}} = \theta_{\text{t}} \left[ {{\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 {\left( {\frac{1}{{\sigma_{\varepsilon }^{2} ({\text{t}})}}} \right)}}}\right.\kern-\nulldelimiterspace} \!\lower0.7ex\hbox{${\left( {\frac{1}{{\sigma_{\varepsilon }^{2} ({\text{t}})}}} \right)}$}}} \right] $$

In this case increases in the variance of the signal concerning earnings simply serve to amplify the response of prices to analyst’s revisions. Now price change conditional on the valuation signal, θ_t, becomes simply the inverse of Bayesian precision of that signal multiplied by the signal’s value itself, θ_t. A rising AIR $\frac{{\sigma_{\varepsilon }^{2} ({\text{t}})}}{{\sigma_{\varphi }^{2} ({\text{t}})}}$ heightens price responses to the signal contained in the forecast or recommendation as its informational content regarding value intensifies. So in this simple “pure” signal case very precise signals intensify the signal’s impact. This is because $\frac{1}{{\sigma_{\varepsilon }^{2} }}$ grows large and so the square bracketed term in the expression expands. Very diffuse signals, for which $\frac{1}{{\sigma_{\varepsilon }^{2} }}$ (earnings news precision) is small, cause impact of the signal on price is greatly muted, or “noisy”, in terms of price response it induces. For example, consider when $\sigma_{\varepsilon }^{2} = \sigma_{\varphi }^{2}$ then the noisy price signal is halved compared to when there is no noise at all. Re-introducing $\sigma_{\varphi }^{2}$ now simply serves to dull the impact of analysts’ forecast revisions since they can no longer be seen as unambiguous signals regarding fundamental/true share value which is solely determined by news about earnings.

1.2 The portfolio variance specification

Here I repeat the exercise presented in Fig. 7 using an alternative specification to Eq. (4) in the main paper. This grosses up the price response not by the level of noise variance, but rather by the portfolio variance of an equally weighted portfolio of assets whose volatility is induced by noise and earnings movements, which I denote as $\sigma_{p}^{2}$.

$$ \begin{gathered} \Updelta {\text{P}}_{\text{t}} = ({\text{P}}_{\text{t}}^{\text{new}} |\theta_{\text{t}} ) - {\text{P}}_{\text{t}} = \left( {\sigma_{\text{p}}^{2} \times \sigma_{{\varphi {\text{t}}}}^{2} } \right)\left[ {{1 \mathord{\left/ {\vphantom {1 {\left( {\frac{{\left( {1 + \sigma_{\varphi }^{2} ({\text{t}})} \right)}}{{\sigma_{\varepsilon }^{2} ({\text{t}})}}} \right)}}} \right. \kern-\nulldelimiterspace} {\left( {\frac{{\left( {1 + \sigma_{\varphi }^{2} ({\text{t}})} \right)}}{{\sigma_{\varepsilon }^{2} ({\text{t}})}}} \right)}}} \right] \hfill \\ {\rm where} \hfill \\ \sigma_{\text{p}}^{2} = \sqrt {x_{\varepsilon }^{2} {\text{SD}}_{\varepsilon }^{2} + x_{\varphi }^{2} {\text{SD}}_{\varphi }^{2} + 2x_{\varepsilon } x_{\varphi } {\text{SD}}_{\varepsilon } {\text{SD}}_{\varphi } \rho_{\varepsilon \varphi } } \hfill \\ \end{gathered} $$

(5)

and x_φ and x_φ are the weights on noisy and earnings induced variance assets. I set x_ε = x_φ = 0.5 in this simple numerical exercise and then ρ_εφ increments of 1% are considered to trace out their impact on the AIR the chosen metric of how well analysts’ advice about value is conveyed to investors. This alternative specification produces a set of comparative statics rather similar to that given in the main paper using Eq. (4). I conclude that a variety of simple intuitive specifications are consistent with my basic argument that increased noise about value intensifies the price impact of bias, although that impact is soon eroded if noise rises above some “critical”(typically fairly low) level. In markets dominated by noise-trader induced variance the value of investment advice itself is increasingly called into question by investors eager to receive only value relevant advice.