Last month, Jonas Heese published a paper on “Government Preferences and SEC Enforcement” which purports to show that the US Securities and Exchange Commission (SEC) refrains from taking enforcement action against companies for accounting restatements when such action could cause large job losses particularly in an election year and particularly in politically important states. The results show that:
- The SEC is less likely to take enforcement action against firms that employ relatively more workers (“labour intensive firms”).
- This effect is stronger in a year in which there is a presidential election
- The election year effect in turn is stronger in the politically important states that determine the electoral outcome.
- Enforcement action is also less likely if the labour intensive firm is headquartered in a district of a senior congressman who serves on a committee that oversees the SEC
All the econometrics appear convincing:
- The data includes all enforcement actions pertaining to accounting restatements over a 30 year period from 1982 to 2012: nearly 700 actions against more than 300 firms.
- A comprehensive set of control variables have been used including the F-score which has been used in previous literature to predict accounting restatements.
- A variety of robustness and sensitivity tests have been used to validate the results
But then, I realized that there is one very big problem with the paper – the definition of labour intensity:
I measure LABOR INTENSITY as the ratio of the firm’s total employees (Compustat item: EMP) scaled by current year’s total average assets. If labor represents a relatively large proportion of the factors of production, i.e., labor relative to capital, the firm employs relatively more employees and therefore, I argue, is less likely to be subject to SEC enforcement actions.
Seriously? I mean, does the author seriously believe that politicians would happily attack a $1 billion company with 10,000 employees (because it has a relatively low labour intensity of 10 employees per $1 million of assets), but would be scared of targeting a $10 million company with 1,000 employees (because it has a relatively high labour intensity of 100 employees per $1 million of assets)? Any politician with such a weird electoral calculus is unlikely to survive for long in politics. (But a paper based on this alleged electoral calculus might even get published!)
I now wonder whether the results are all due to data mining. Hundreds of researchers are trying many things: they are choosing different subsets of SEC enforcement actions (say accounting restatements), they are selecting different subsets of companies (say non financial companies) and then they are trying many different ratios (say employees to assets). Most of these studies go nowhere, but a tiny minority produce significant results and they are the ones that we get to read.
Thu, 22 Jan 2015
In high frequency trading, nine minutes is an eternity: it is half a million milliseconds – enough time for five billion quotes to arrive in the hyperactive US equity options market at its peak rate. On a human time scale, nine minutes is enough time to watch two average online content videos.
So what puzzles me about the soaring Swiss franc last week (January 15) is not that it rose so much, nor that it massively overshot its fair level, but that the initial rise took so long. Here is the time line of how the franc moved:
- At 9:30 am GMT, the Swiss National Bank (SNB) announced that it was “discontinuing the minimum exchange rate of CHF 1.20 per euro” that it had set three years earlier. I am taking the time stamp of 9:30 GMT from the “dc-date” field in the RSS feed of the SNB which reads “2015-01-15T10:30:00+01:00” (10:30 am local time which is one hour ahead of GMT).
- The head line “SNB ENDS MINIMUM EXCHANGE RATE” appeared on Bloomberg terminals at 9:30 am GMT itself. Bloomberg presumably runs a super fast version of “if this then that”. (It took Bloomberg nine minutes to produce a human written story about the development, but anybody who needs a human written story to interpret that headline has no business trading currencies).
- At the end of the first minute, the euro had traded down to only 1.15 francs, at the end of the third minute, the euro still traded above 1.10. The next couple of minutes saw a lot of volatility with the euro falling below 1.05 and recovering to 1.15. At the end of minute 09:35, the euro again dropped below 1.05 and started trending down. It was only around 09:39 that it fell below 1.00. It is these nine minutes (half a million milliseconds) that I find puzzling.
- The euro hit its low (0.85 francs) at 09:49, nineteen minutes (1.1 million milliseconds) after the announcement. This overshooting is understandable because the surge in the franc would have triggered many stop loss orders and also knocked many barrier options.
- Between 09:49 and 09:55, the euro recovered from its low and after that it traded between 1.00 and 1.05 francs.
It appears puzzling to me that no human trader was taking out every euro bid in sight at around 9:33 am or so. I find it hard to believe that somebody like a George Soros in his heyday would have taken more than a couple of minutes to conclude that the euro would drop well below 1.00. It would then make sense to simply lift every euro bid above 1.00 and then wait for the point of maximum panic to buy the euros back.
Is it that high frequency trading has displaced so many human traders that there are too few humans left who can trade boldly when the algorithms shut down? Or are we in a post crisis era of mediocrity in the world of finance?
Updated to correct 9:03 to 9:33, change eight billion to five billion and end the penultimate sentence with a question mark.
Tue, 13 Jan 2015
Two months back, I wrote a blog post on how the Basel Committee on Payments and Market Infrastructures was reckless in insisting on a two hour recovery time even from severe cyber attacks.
I think that extending the business continuity resumption time target to a cyber attack is reckless and irresponsible because it ignores Principle 16 which requires an FMI to “safeguard its participants’ assets and minimise the risk of loss on and delay in access to these assets.” In a cyber attack, the primary focus should be on protecting participants’ assets by mitigating the risk of data loss and fraudulent transfer of assets. In the case of a serious cyber attack, this principle would argue for a more cautious approach which would resume operations only after ensuring that the risk of loss of participants’ assets has been dealt with. ... The risk is that payment and settlement systems in their haste to comply with the Basel mandates would ignore security threats that have not been fully neutralized and expose their participants’ assets to unnecessary risk. ... This issue is all the more important for countries like India whose enemies and rivals include some powerful nation states with proven cyber capabilities.
I am glad that last month, the Reserve Bank of India (RBI) addressed this issue in its Financial Stability Report. Of course, as a regulator, the RBI uses far more polite words than a blogger like me, but it raises almost the same concerns (para 3.58):
One of the clauses 31 under PFMIs requires that an FMI operator’s business continuity plans must ‘be designed to ensure that critical information technology (IT) systems can resume operations within two hours following disruptive events’ and that there can be ‘complete settlement’ of transactions ‘by the end of the day of the disruption, even in the case of extreme circumstances’. However, a rush to comply with this requirement may compromise the quality and completeness of the analysis of causes and far-reaching effects of any disruption. Restoring all the critical elements of the system may not be practically feasible in the event of a large-scale ‘cyber attack’ of a serious nature on a country’s financial and other types of information network infrastructures. This may also be in conflict with Principle 16 of PFMIs which requires an FMI to safeguard the assets of its participants and minimise the risk of loss, as in the event of a cyber attack priority may need to be given to avoid loss, theft or fraudulent transfer of data related to financial assets and transactions.
Sat, 03 Jan 2015
I read two papers last week that introduced heterogeneous investors into multi factor asset pricing models. The papers help produce a better understanding of momentum and value but they seem to raise as many questions as they answer. The easier paper is A Tug of War: Overnight Versus Intraday Expected Returns by Dong Lou, Christopher Polk, and Spyros Skouras. They show that:
100% of the abnormal returns on momentum strategies occur overnight; in stark contrast, the average intraday component of momentum profits is economically and statistically insignificant. ... In stark contrast, the profits on size and value ... occur entirely intraday; on average, the overnight components of the profits on these two strategies are economically and statistically insignificant.
The paper also presents some evidence that “is consistent with the notion that institutions tend to trade intraday while individuals are more likely to trade overnight.” In my view, their evidence is suggestive but by no means compelling. The authors also claim that individuals trade with momentum while institutions trade against it. If momentum is not a risk factor but a free lunch, then this would imply that individuals are smart investors.
The NBER working paper (Capital Share Risk and Shareholder Heterogeneity in U.S. Stock Pricing) by Martin Lettau, Sydney C. Ludvigson and Sai Ma presents a more complex story. They claim that rich investors (those in the highest deciles of the wealth distribution) invest disproportionately in value stocks, while those in lower wealth deciles invest more in momentum stocks. They then examine what happens to the two classes of investors when there is a shift in the share of income in the economy going to capital as opposed to labour. Richer investors derive most of their income from capital and an increase in the capital share benefits them. On the other hand, investors from lower deciles of wealth derive most of their income from labour and an increase in the capital share hurts them.
Finally, the authors show very strong empirical evidence that the value factor is positively correlated with the capital share while momentum is negatively correlated. This would produce a risk based explanation of both factors. Value stocks lose money when the capital share is moving against the rich investors who invest in value and therefore these stocks must earn a risk premium. Similarly, momentum stocks lose money when the capital share is moving against the poor investors who invest in momentum and therefore these stocks must also earn a risk premium.
The different portfolio choices of the rich and the poor is plausible but not backed by any firm data. The direction of causality may well be in the opposite direction: Warren Buffet became rich by buying value stocks; he did not invest in value because he was rich.
But the more serious problem with their story is that it implies that both rich and poor investors are irrational in opposite ways. If their story is correct, then the rich must invest in momentum stocks to hedge capital share risk. For the same reason, the poor should invest in value stocks. In an efficient market, investors should not earn a risk premium for stupid portfolio choices. (Even in a world of homogeneous investors, it is well known that a combination of value and momentum has a better risk-return profile than either by itself: see for example, Asness, C. S., Moskowitz, T. J. and Pedersen, L. H. (2013), Value and Momentum Everywhere. The Journal of Finance, 68: 929-985)