Covid-19 and Corporate India

by Aakriti Mathur and Rajeswari Sengupta.

India is dealing with a massive shock in the form of the Covid-19 pandemic. The first case was reported in India on 30 January, 2020. By middle of March the disease had begun spreading rapidly across the country. To prevent the spread of the virus the Indian government announced a nationwide lockdown on 24 March. The pandemic and the lockdown affected nearly all firms and sectors of the economy; however, there are likely to be significant heterogeneities.

We propose a novel approach to identify firms that may have had greater exposure to the pandemic even before it assumed serious proportions in India, by virtue of, for example, their connections to other affected countries, among others. These firms may have fared worse when the lockdown was announced. We also examine the pre-pandemic balance sheet characteristics that may have worsened the impact of the lockdown on some firms compared to the others.

Analysing earnings call reports

We propose the use of earnings call transcripts as an important source of information for gauging a firm's fundamental exposure to the pandemic.

Earnings calls typically follow the presentation of a firm's quarterly results. These calls are attended by senior management of the firm (for example, the CEO, CFO, MD, etc), who present short prepared remarks, and then open the floor to questions from analysts. This implies that the calls are more spontaneous as compared to say the firm's annual report, because the senior management answers questions on the fly from the audience. These reports therefore convey not just fundamental financial information, but also analysts' and managers' opinion about the firm (Borochin et al., 2018).

A significant part of the literature focuses on the tone and sentiment of these reports and their implications for stock market returns, trading volumes (Frankel et al., 1999; Bushee et al., 2003, 2004; Brown et al., 2004), and options pricing (Borochin et al., 2018). Our work closely relates to two recent papers, Hassan et al (2020) and Ramelli and Wagner (2020). Both these studies use the information contained in earnings call transcripts. Hassan et al (2020) focus on globally listed firms, and study whether firms that were more exposed to previous disease outbreaks such as SARS and MERS were better prepared for the 2020 pandemic, and therefore had higher equity returns, than those who were not. Ramelli and Wagner (2020) analyse characteristics of US firms that explain both their stock market performance between January and March 2020 and their discussions of Covid-19 in the earnings transcripts.

Unlike these papers, we use the information in earnings call reports, to measure fundamental exposure of Indian firms to the pandemic. We are interested in using this information to study the equity market performance of the firms around the largest, most stringent lockdown announced in the world (at the time). Our analysis complements earlier work by Sane and Sharma (2020) who calculated the liquidity cover of listed firms in India in the face of large revenue shocks during the pandemic and Bansal et al. (2020) who also examine variations in the market valuation of firms on account of firm-specific characteristics during the pandemic. In this work, we take a more holistic view of firm-level vulnerabilities, examining the fundamental exposure to the pandemic, as well as the role of financial flexibilities, including liquidity. We also focus on one specific event -- the 24 March lockdown -- in order to obtain greater precision.

We focus on earnings calls conducted by firms in January and February 2020, when the case load was still low in India but the pandemic had begun spreading in other countries. These are calls discussing the income statements of October-December, 2019 (Q3 FY20) and January-March, 2020 (Q4 FY20) respectively, of the Indian financial year.

When India reported its first case of the Covid-19 pandemic on 30 January, 2020, close to 7,700 people had been infected all over the world, the majority being in China. Other countries such as the US, Australia, Germany, Japan, South Korea, UAE and HongKong had started reporting Covid-19 cases. By end February, it had morphed into a full blown public health crisis. The total number of infections globally had risen to more than 83,000, with a death toll of more than 2,800. While the disease was spreading rapidly in countries such as Italy, South Korea, France, the US and Iran, these were still early days of the pandemic in India which had less than 10 confirmed cases.

Focusing on the call reports of Jan-Feb 2020 enables us to analyse the firms' exposure to the pandemic at a time when the disease was still at a nascent stage in India unlike say March when the spread of the pandemic had begun affecting most firms. It also allows for easier identification of firm exposure because it is not muddled by domestic policy interventions. For example, there were only 13 Covid-19 related notifications issued by the Indian government in February, compared to 266 in March, as listed by PRS Legislative Research. Hence, from March onwards, the stock market performance of all firms was likely to be affected by these interventions over and above firm-specific concerns around the disease itself.

We start with a sample of the largest listed firms on the Nifty500 index of the National Stock Exchange (NSE) of India. Of the 500 firms in the index, we have access to the call reports of 196 firms in January-February 2020, and of 90 firms in April-May 2020.

Which firms had exposure to Covid-19 in Jan-Feb 2020?

We interpret the number of times a firm mentioned Covid-19 related words in its call reports as an indicator of its exposure to the pandemic. Accordingly, we count the number of times Covid-19 and related words (such as "coronavirus", "pandemic", "ncov", "sarscov", "epidemic" etc) are mentioned in the quarterly earnings call reports of the 196 firms in January-February 2020, and also of the 90 firms in April-May 2020 for the sake of comparison. We briefly summarise our findings below.

• Only one-third of the firms in our Jan-Feb sample mention Covid-19 or related words. The average number of times these words are mentioned is three. Only three of the firms discussing the pandemic are in the financial services sector.
• All 90 firms in the Apr-May 2020 sample mention Covid-19 or related words, demonstrating the extensive spread of the disease by this time. The average occurrence of the words per report is ten times higher, close to 31. This reflects our earlier concern that from March onwards, all firms had become exposed to the pandemic.
• Even in Jan-Feb 2020, there were sector-wise heterogeneities in Covid-19 discussions, as shown in figure 1 below.
• The occurrences of Covid-19 related words were higher in those sectors which presumably have more fundamental exposure to the pandemic, for example in the form of supply-chains with China or other early-affected countries. Some of these sectors are pharmaceuticals, consumer goods, automobile, chemicals etc.
• Firms in health care services, financial services, media and entertainment, power and telecom industries either did not mention or mentioned much less pandemic related words during this period in their call reports. With the possible exception of health care, these sectors were likely to be affected due to indirect exposure to the pandemic.

Figure 1: Sector-wise occurrences of Covid-19 related words in Jan-Feb call reports

We also look at the firms that mentioned "supply", "demand" and "uncertainty" related words in context of the Covid-19 discussion in their call reports. These are likely to be the most common channels of disruption faced by the firms during the pandemic. In results not reported here we find that firms in sectors with higher than average mentions of Covid-19 related words also had higher than average mentions of "supply" related words in the sentences where Covid-19 was discussed. For firms in the services sector, mentions of "demand" related words in the context of the disease were higher. For all the sectors having higher than average mentions of Covid-19 related words, we also find significantly higher mentions of "uncertainty" related words in the context of the pandemic.

This preliminary analysis gives us an idea of which firms and sectors had greater exposure to the pandemic as early as Jan-Feb 2020 when the disease still hadn't spread in India.

For subsequent analysis, we consider the firms that mention Covid-19 in Jan-Feb 2020 as our "treated" sample and those that did not discuss the pandemic as our "control" sample. A relevant question to ask is how similar are the "treated" and "control" samples in terms of their key balance sheet characteristics. Using annual data from the pre-pandemic period (ending in March 2019) from the Prowess database of CMIE, we compare the two sets of firms in size, age, profit, foreign exchange earnings, inventories, cash balances etc. For ease of comparison, we drop the three firms that are in the financial services sector and that mentioned pandemic related words in the Jan-Feb call reports. As shown in table 1 below, we do not find any major difference between these two groups of firms, except that the "treated" firms are on average older and hold higher inventories than the "control" group firms.

Table 1: Summary statistics for non-financial firms: Data as of March 31, 2019

	No. of firms with no COVID mentions in Jan-Feb 2020	No. of firms with COVID mentions in Jan-Feb 2020
	96	60
Variable	Mean of firms with no COVID mentions	Mean of firms with COVID mentions
Age	35	43.7
Log Size	11.19	11.13
Leverage (Debt/Assets)	0.15	0.16
PBDITA/Total Sales	0.26	0.23
FX Earnings/Total income	0.30	0.27
Cash and Bank balance/Total Assets	0.07	0.05
Trade Receivables/Total Assets	0.14	0.13
Inventories/Total Assets	0.09	0.12
Operating Expenses/Total Income	0.76	0.78

What kind of exposure did firms have to the pandemic in Jan-Feb, 2020?

We next analyse the context within which the firms discussed the pandemic in their call reports, for example, references to supply-chains, demand disruptions, or uncertainty due to the pandemic and so on. This will give us a sense of the kind of exposure the firms may have had to the pandemic in the early part of 2020.

We use the techniques applied in Mathur and Sengupta (2019). For every firm's call report, we first isolate the sentences that contain Covid-19 related words. There are 176 sentences in total for the Jan-Feb 2020 reports. Then, we create a word cloud with the most frequently occurring words in these sentences, after stripping out stop-words and other uninformative words.

The word cloud for Q3 FY19-20 reports is shown in figure 2. The size of each word is directly proportional to its frequency in the sentences. We plot the 50 most frequently occurring words. All the coronavirus related words, which are the most common words in these sentences by construction, are not plotted here for ease of comprehension.

• The words "china" and "impact" occur most frequently indicating that firms were talking about the origin of the coronovirus disease and its effect.
• We also see words related to areas where the impact of the pandemic was potentially anticipated or the expected transmission channels of the disease such as "earnings", "shipping", "pharma", "macro", "supply chain", "trade", "logistics", "imports", "demand", "supply", "prices" etc.

Figure 2: Word clouds of sentences with Covid-19 related words in Jan-Feb call reports

Which firms were more affected by the 24 March lockdown announcement?

The 24 March lockdown in India was regarded as one of the most severe lockdowns in the world, based on data from the Oxford COVID-19 Government Response Tracker. All transport services, except those for essential personnel, were suspended, in addition to all educational, commercial, and private establishments (see here). The lockdown affected all sectors of the Indian economy. The stock market reacted negatively overall. This is not surprising, since stock prices reflect changes in expected future cash flows and/or discount rates. However it is possible that some firms were more affected than the others depending on their exposure to the pandemic as well as pre-pandemic characteristics.

To measure the differential, cross-sectional responses of firms to the lockdown announcement, we use high-frequency stock market data and an event study methodology. We have two main hypotheses.

Our primary hypothesis is that firms that were more exposed to the pandemic and mentioned Covid-19 in their earnings call reports in Jan-Feb 2020 (the "treated" group) fared worse than the "control" group when the lockdown was announced.

• If investors believe that firms who discussed Covid-19 and its implications for their businesses early on in the year are more exposed to the virus, for example due to supply chains with China, or factories in badly-affected countries like Italy, then they would revise their expectations of future profitability downwards in response to the lockdown. Therefore, we would see that treated firms as a whole perform worse than control firms.
• If investors believe that early discussions of the pandemic implied that these firms were better prepared to weather the storm, then their returns would be better than those that seemed to have been caught "off-guard". We hypothesise that the former is likelier than the latter, since it is not clear how firms could have unilaterally prepared for the over-arching extent of the shock (such as to demand disruptions) just a couple of months in advance.

Our second hypothesis is that low-profitability firms with higher share of foreign exchange earnings, higher share of inventories, greater dependence on trade credit and higher operating expenses should have witnessed lower stock market returns when the lockdown was announced, compared to more domestically oriented firms which were more profitable, were holding lower inventories, had lower dependence on trade credit and also lower operating expenses.

Estimation strategy

We use a difference-in-difference strategy to estimate the impact of the lockdown event on firms' stock market returns. We consider all "treated" firms as one group by using a dummy ("Covid dummy"). Our dependent variable is the cumulative abnormal stock market returns (CARs) for each firm over a window of (-1, +2) days around the lockdown event, i.e. between 23 March (Monday) and 26 March (Thursday). To obtain these abnormal returns, we estimate a market model (i.e. controlling for movements in the Nifty50 index), as shown in the equation below, over a period of 81 days prior to March 24. More specifically our window starts 91 days prior to the lockdown and stops 11 days before the lockdown. The model specification is:

$$\text{Daily firm returns}_{firm,t} = \alpha + \beta~\text{Daily Nifty50 returns}_{t} + \epsilon$$

The advantage of using a tight window around the event is that it better accounts for anticipation effects and other confounding factors. We use a cross-sectional ordinary least squares regression shown in the equation below, to regress the firm-specific CARs on the "Covid dummy" and on a host of balance sheet variables. Among the regressors, of particular interest is the "Covid dummy" which tells us the difference in CARs between the "treated" and the "control" firms. Other regressors include the balance sheet variables shown in table 1 above as well as dummy variables for the sectors that the firms belong to. Firm level annual balance sheet variables are as of March 31, 2019.

$$\text{CARs around event window}_{firm} = \alpha_{sector} + \beta~\text{Covid Dummy}_{firm} + \\ \log(age)_{firm} + \log(size)_{firm} + Controls_{firm} + \epsilon $$

Results

We summarise our results in Figure 3. In panel (a) we plot the results from our baseline model (model 1) which includes only age and size of firms, and the sector dummies, over and above the Covid dummy. We also plot the results from models 2 to 6 where in addition to the Covid dummy, age, size, and sectors, we sequentially add the regressors of interest: profit, FX earnings, inventories, operating expenses, and trade receivables. We also investigate the role of cash, leverage, borrowing composition, and tangible assets. Here we only report results that are significant at 90% confidence interval. We list our main findings below.

• In all our specifications, the stock returns of "treated" firms, i.e. those that mentioned Covid-19 in their call reports early on in 2020, significantly underperform (at the 90% confidence level or more) the "control" firms. On average, returns of the "treated" firms are roughly 3.5 percentage points lower.
• We find that the equity returns of more profitable firms outperformed those of less profitable ones by 9 percentage points (model 3). Higher profitability implies higher ability to withstand large revenue shortfalls.
• Firms with a higher share of foreign exchange earnings in their total income performed worse. They were likely to be more affected due to supply and demand disruptions in the rest of the world.
• Firms with high inventories saw 24 percentage points lower returns. High share of inventories in total assets might make it difficult for firms to get rid of their inventories once an economywide lockdown is announced yet they would have had to incur the costs of maintaining these inventories which makes them worse off than firms with lower share of inventories (Banerjee et al., 2020).
• Firms with higher pre-pandemic trade credit reliance saw significantly lower abnormal returns. This is likely because in a broad-based crisis such as this one, credit markets are likely to freeze along both extensive and intensive margins. Thus, rolling over existing trade credit as well as obtaining new supply of trade credit would be difficult. (Banerjee et al., 2020).
• Firms with higher pre-pandemic operating expenses also fared worse once the lockdown was announced. Operating expenses are typically short term expenses. In absence of steady revenues in a lockdown, firms would depend on credit from the financial system to meet these expenses. During a crisis if the financial system is unwilling to offer short term credit (Sengupta and Vardhan, 2020), then these firms are likely to witness lower stock returns.

In figure 3, panel (b), we plot the coefficients on the sector dummies from the baseline model 1, with only age and size included as controls. Automobiles is the benchmark sector. We find that stock returns of more consumer facing sectors (textiles, media and entertainment) and those that rely on supply chains (metals, and oil and gas) did particularly badly when the lockdown was announced. On the other hand, healthcare services in particular outperformed as compared to automobiles.

Figure 3, panel (A): Explaining cumulative abnormal returns around first lockdown (24 March, 2020)

Figure 3, panel (B): Sector dummies from baseline regression

In a nutshell, we find that when the nationwide lockdown was announced on 24 March, firms who mentioned Covid-19 in their earnings calls in early-2020 and hence were more exposed to the pandemic, fared worse than firms who did not discuss the pandemic. This result holds when we account for the sectors and key balance sheet characteristics of the firms.

As discussed in Fahlenbrach et al.(2020), less financially flexible firms are less able to withstand large negative shocks to their revenues, which translates to worse equity market performance. Lower cash, lower profitability, lower diversification in earnings (e.g. higher reliance on foreign exchange revenues) or in borrowing sources (e.g. higher reliance on trade credit) can all be considered indicators of low financial flexibility. In other words, firms that had lower financial flexibility in the pre-pandemic period were worse affected when the lockdown was announced.

We further find that controlling for mentions of "supply" and "demand" related words (not shown here) in the firms' call reports -- which may account for the nature of their exposure to the pandemic -- does not change the results qualitatively, and makes them stronger in some specifications.

Firms with more cash holdings reported higher returns on average around the lockdown announcement, but this effect is not significant (hence, not reported here). In further tests, we find some evidence of non-linearities. Firms with above-median cash holdings significantly outperform their counterpart.

Conclusion

Using the informational content of earnings call reports of some of the largest, non-financial firms in India we throw light on the firms and sectors that may have been more exposed to the pandemic as early as January and February 2020 when as per the official statistics, the disease had still not spread in India. We find that these firms were also worse affected by the announcement of a nationwide lockdown in March compared to firms that were presumably less exposed to the pandemic early on.

Our results highlight the kind of firms that are likely to be more affected when a crisis such as the ongoing one hits the economy. Firms with lower profits, higher share of foreign exchange earnings, higher share of inventories, greater dependence on trade credit and higher operating expenses fared worse on the stock market when the lockdown was announced.

References

Banerjee, R., Illes, A., Kharroubi, E., and Serene, JM. (2020). COVID-19 and corporate sector liquidity, BIS Bulletin No. 10, April, 2020.

Bansal, A., Gopalakrishnan B., Jacob, J., and Srivastava, Pranjal. (2020). When the Market Went Viral: COVID-19, Stock Returns, and Firm Characteristics,

Available at SSRN (June 21, 2020).

Borochin, P.A., Cicon, J.E., DeLisle, R.J., and Price, S.M. (2018). The effects of conference call tones on market perceptions of value uncertainty, Journal of Financial Markets, 40(2018), April, 2018, pp.75--91.

Bushee, B.J., Matsumoto, D.A., Miller, G.S., 2003. Open versus closed conference calls: The determinants and effects of broadening access to disclosure,. Journal of Accounting and Economics, 34(1-3), January, 2003, pp.149--180.

Bushee, B.J., Matsumoto, D.A., Miller, G.S., 2004. Managerial and investor responses to disclosure regulation: the case of Reg FD and conference calls, . The Accounting Review, 79(3), July, 2004, pp.617--643.

Fahlenbrach, R., Rageth K., and Stulz, R. (2020). How valuable is financial flexibility when revenue stops? Evidence from the COVID-19 crisis, NBER Working Papers 27106, May, 2020.

Frankel, R., Marilyn J., and Douglas, S. (2020). An empirical examination of conference calls as a voluntary disclosure medium, Journal of Accounting Research, 37(1), Spring, 1999, pp.133--150.

Hale, T., Webster, S., Petherick, A., Phillips, T., and Kira, B. (2020). Oxford COVID-19 Government Response Tracker, Blavatnik School of Government.

Hassan, A, T. et al (2020). Firm-level exposure to epidemic diseases: COVID-19, SARS, and H1N1, NBER Working Papers 26971, April, 2020.

Mathur, A., and Sengupta, R. (2019). Analysing monetary policy statements of the Reserve Bank of India, IHEID Working Papers 08-2019, May, 2019.

Ramelli, S., and Wagner, A.F. (2019). Feverish stock price reactions to COVID-19, Swiss Finance Institute Research Paper No. 20-12, Forthcoming Review of Corporate Finance Studies, March, 2020.

Sane, R., and Sharma, A. (2020). Holding their breath: Indian firms in an interruption of revenue, The Leap Blog, 03 April, 2020.

Sengupta R., and Vardhan, H. (2020). Policymaking at a time of high risk-aversion Ideas for India, 06 April, 2020.

Aakriti Mathur is a PhD candidate at The Graduate Institute (IHEID), Geneva. Rajeswari Sengupta is an Assistant Professor of Economics at IGIDR, Mumbai.

Delays in liquidated and resolved firms: Visualisation of an output measure of the Indian bankruptcy reform

by Geetika Palta, Anjali Sharma, Susan Thomas.

The ultimate objective of the Indian bankruptcy reform is to get up to plausible recovery rates and change the behaviour of borrowers. The key tool for achieving these objectives is reducing the delay. In the existing literature, we know that there are large delays, particularly for large firms (Bhatia et. al. 2019, Felman et. al. 2019, Shah 2018). The most important proximate objective of the Indian bankruptcy reform is to reduce delays in the bankruptcy process (Shah and Thomas, 2018).

A great deal of the focus so far has been upon the average value of the delay. It is, however, important to look at the full distribution of the delay, and not just the sample mean. Box-and-whisker plots are a nice visualisation tool through which we can see more than just the sample mean. In this article, we (a) Construct a visualisation of a key output measure for the Indian bankruptcy reform : a box-and-whisker plot for the delay associated with Resolved and Liquidated firms; and (b) Argue that this output measure is likely to get worse in coming days.

The overall distribution of the delay

Sometimes, it is argued that the right way to measure the delay is to exclude certain elements of the delay, which is not correct seen from first principles. The fundamental fact about distressed firms is that every day of delay reduces recovery rates and hampers economic dynamism. For an analogy, a sick animal is unproductive and suffers, regardless of whether the vet takes the weekend off or not.

We work with two years of data about cases that have concluded and exited from the IRP. These are obtained from the IBBI website. In this data, 1383 cases embarked into the Insolvency Resolution Process (IRP) from January 2017 to December 2018. Of these, 79 concluded with an accepted resolution plan and 304 cases that concluded with the firm being put into liquidation. This yields the following box-and-whisker plots:

Figure 1: Box-whisker plots for the delay of IBC cases, under three buckets (Ongoing, Liquidated or Resolved)

Let's start at the right column (for resolved cases). The bottom pane shows that 79 firms were resolved. The upper pane depicts the range of values for these firms. The black horizontal line is the median, and the box is drawn from the 25th to the 75th percentile values for the delay. The dots show the most extreme values. A key finding here is that the median resolved case took more than 270 days (the horizontal red line).

When we look at the liquidated cases, things are slightly better. The black line -- the median delay -- is close to 270 days. It still says that half of liquidated cases took more time than the legal limit of 270 days. A little under 25 percent of the liquidation cases reached their conclusion in 180 days, while very few of the resolved cases concluded within 180 days. But more than 50 percent of the liquidation cases concluded within the 270 days limit, while a little more than 25 percent of the resolved cases were done by this time.

These two pictures -- the box-and-whisker plots for resolved and liquidated cases -- are a nice visualisation of a key output measure of the Indian bankruptcy reform. The trouble is, so far, we have seen only 79 + 304 cases reach the conclusion. These statistical estimates are censored: the cases that have finished are likely to be the ones where the IBC fared relatively well. The bulk of the action is in the Ongoing cases, and there are over 900 of them. For these, the median delay is already in the region of 270 days.

The box-and-whisker plots for Liquidated and Resolved cases, which is the output measure of the Indian bankruptcy reform, will be modified in the future based on cases emerging out of the Ongoing bucket. Very crudely, we may conjecture that if all the ongoing cases finish tomorrow, the 25th and 50th and 75th percentile values of the overall distribution will be much like those seen as of today with the Liquidated and Resolved cases. But this is an over-optimistic scenario. In fact, cases will only trickle out in the future with higher delays, cases where the median delay has already reached about 270 days. Therefore, as cases emerge out of the Ongoing bucket in the future, the box-and-whisker plots for Liquidated and Resolved cases are going to get worse.

How might the output measure evolve in the future?

The most interesting question before us is: In the future, when Ongoing cases trickle out into completion, how will the output measures (the box-and-whisker plots of Resolved and Liquidated cases) shape up?

To help visualise what comes next, we create the box-and-whisker plots for the Ongoing cases by quarter, from Q1 (Jan to Mar) 2017 to Q4 (Oct to Dec) 2018 in Figure 2. The $x$ axis shows the quarter in which cases were admitted into the IRP. The top pane of the graph shows the box-and-whisker plot for the days in IRP for the Ongoing cases only (those which have not concluded as of Dec 2018) and the bottom pane shows the number of firms.

The graph for the number of firms shows that a large fraction of the Ongoing cases have started their IRP in the last three quarters of 2018 -- between April to December 2018. Among these three quarters, there is a near split of about 33%-33%-33%, between cases that have spent more than 270 days, between 180 and 270 days, and below 180 days.

Figure 2: Delays associated with ongoing cases, organised by quarter

To some extent, these results are mechanically driven by the facts of time. But the results are remarkable nonetheless. As an example, the (few) pending cases from Q1 2017 have already spent over 700 days of delay! When these cases complete, they will push the outcome measures in an adverse direction.

On the other hand, a good number of cases are in 2018 where, so far, the delay that has been clocked is relatively low. If, hypothetically, the Indian bankruptcy reform suddenly works better, then a slew of cases can complete, and then the outcome measure may even improve.

Conclusions

1. The box-and-whisker plot of the delay (measured in calendar days) for Resolved and Liquidated firms is a nice visualisation of a key output of the Indian bankruptcy reform.
   
   
2. It shows a gloomy picture, where over half of the delays are worse than the outer limit in the law of 270 days.
   
   
3. Looking into the future, based on the delays already incurred with Ongoing cases, the output measure is likely to get worse.

References

 

Time to resolve insolvencies in India, Surbhi Bhatia, Manish Singh, Bhargavi Zaveri, The Leap Blog, 11 March 2019.

The RBI-12 cases under the IBC by Josh Felman, Varun Marwah, Anjali Sharma, 2019 (forthcoming).

Sequencing issues in building jurisprudence: the problems of large bankruptcy cases, Ajay Shah, The Leap Blog, 7 July 2018.

The Indian bankruptcy reform: The state of the art, 2018, Ajay Shah, Susan Thomas, The Leap Blog, 22 December 2018.


The authors are researchers at the Indira Gandhi Institute for Development Research.

Watching markets work: The dramatic events of 8 November 2016

by Anurag Dutt, Sargam Jain, Ajay Shah, Susan Thomas.

On November 9th, 2016, Indian financial markets were asked to digest two major events. At 8:30 PM on 8th evening, the announcement had come out about de-monetisation of the 500/1000 rupee notes. And, by late night on the 8th, there was news of Donald Trump's early gains in the U.S. presidential elections. At 9:28 a.m. IST on the 9th, Donald Trump won Florida. At this point in time, for Hillary Clinton to win, she had to win the states of Wisconsin and Michigan, where Trump was already leading. At 1:00 p.m. IST, Clinton conceded.

The two events contain an interesting contrast. With the US presidential election, the betting markets were reporting a 20% chance of a Trump win. But de-monetisation was on nobody's radar. It was not part of our distribution.

In this article, we go back to those events and look at how the Indian financial markets responded to the major events.

Prices

Figure 1: Prices

All the graphs in this article show physical time from the morning of 7 November to the end of 11 November. The top panel shows the equity market, with Nifty spot in blue and Nifty futures in grey. On 8th evening, Nifty rose slightly, which suggests that there was no insider trading based on the de-monetisation, and nobody had a sense that Trump would win. On 9th morning, Nifty opened sharply down, reflecting both elements of news. By the time Nifty trading stopped, Nifty had remarkably come back to 8,432. With the benefit of hindsight, we know that the market was too quick in reconciling itself to the news. By the end of the week, Nifty had fallen further. On 2 December, Nifty closed at 8086, showing that the de-monetisation news had not been fully understood even by 11 November.

We may speculate that in the Indian equity market, there is a lot of focus on information production about individual stocks, but low capabilities in macroeconomics. In many previous events also, we have seen the market being relatively slow in understanding far-reaching macroeconomic developments.

The middle panel is the USD/INR. The blue line is USD/INR futures (which trade 09:00 a.m. to 05:00 p.m.) and the grey line is the spot market, which can trade for 24 hours a day. Let's think about the evening of 8th. At 8:30 PM, there was the de-monetisation announcement. Remarkably, the spot market showed an appreciation of 40 paisa. This shows that the currency traders of 8th evening did not understand the de-monetisation. In the late night, news of Trump's success started trickling in. In the morning of 9th, INR depreciated reflecting both elements of news. In the Indian afternoon, the USD moved sharply when Clinton conceded, which gave an appreciation.

For the rest of the week, INR depreciated as the bad news sank in. One element that was at work was the huge demand in India to convert 500/1000 rupee notes into US dollars. Anecdotal reports suggest that by 10th, the entire inventory of US dollars in the Indian black market had been exhausted. This would have triggered off demand for dollars, and fueled the depreciation. To the credit of RBI, they let the market do its job; they did not interfere with a large INR depreciation.

The bottom panel is gold. The blue line is MCX gold futures and the grey line is the CME gold futures. MCX gold trading enjoys long hours: from 10:00 a.m. to 11:30 p.m. In the late hours of 8th November, gold prices became very volatile when the de-monetisation announcement came in. Trump's lead in the U.S elections inflated gold prices when the market opened on 9th.

There were two distinct things going on. Worldwide, buying gold is a vote of no-confidence in civilisation and paper money. The Trump win would have encouraged many people worldwide to shift their holdings in favour of more gold. In India, the de-monetisation announcement had an abstract implication (mistrust of the Indian rupee, mistrust of the Indian State) which would have encouraged a higher weightage of gold in the portfolio, but there was also an immediate and practical dimension. Thousands of people flocked to Sarafa bazaars to exchange their high denomination notes for gold in the morning of 9th November.

These extremes were somewhat unwound in the rest of the week. Perhaps there was a premium on physical gold available in India on the 9th. Within a few hours, gold bars could be flown in from Dubai and Singapore, through which the Indian spot price would come back to the world price. But on the day of the 9th, there was extreme demand for gold and the Indian price deviated from the world price.

Turnover

Figure 2: Turnover

Turnover on equity derivatives -- Nifty futures and Nifty ATM options -- is a critical element of Indian price discovery. On the 9th, markets opened with very high trading intensity in both Nifty futures and options. The turnover in near month at-the-money options surged again when Trump's victory was confirmed. There is an interesting pattern thereafter: On 10th and 11th, the futures activity was larger. This runs against the normal pattern in India, where options trading is favoured owing to the high securities transaction tax (STT) on futures trading. This is worth exploring further. Perhaps this odd behaviour is being induced by errors in the rules for initial margin calculation.

In the currency market, futures trading dominates as there is no STT-related distortion. Here, we see we got a big surge of trading on 9th morning, a smaller surge in the afternoon when Clinton conceded, and then a fading away of excess turnover through the rest of the week.

MCX was open for business when the de-monetisation announcement came out, and reaped a bonanza with a massive increase in turnover on 8th evening. For most financial traders in India, on 8th evening, MCX was the only game in town as NSE and BSE were closed. From 9th onwards, the patterns in turnover are similar to those seen with the other two markets: a surge on 9th morning, a surge when Clinton conceded, and a gradual phaseout of extraordinary turnover through the week.

Market efficiency

Figure 3: Violations of no-arbitrage on the futures market

On 9th morning, when NSE opened for trading, there was a huge mispricing between the Nifty spot and the Nifty futures. Once that was corrected, for the rest of the week, pricing errors were comparable to those seen before the news, but the basis risk was higher.

The pricing errors on USD/INR futures are surprisingly small when compared with those seen on the equity market. This suggests there is ample capital in currency futures arbitrage, relative to the small size of the market.

With gold, what we are reporting is the pricing error between the MCX gold futures and the CME gold futures. As emphasised earlier, there was a large dislocation on the Indian gold spot market on 9th, as many people were buying gold. The Indian gold spot price fell out of sync with the world gold price. This is showing up as large pricing errors in the bottom panel, and normalcy is attained as enough planes land in India bearing physical gold.

Figure 4: Violations of put-call parity on the options market

Put-call parity held up pretty well through these events, for both Nifty and USD/INR. A brief large error was found on the morning of the 9th, on the Nifty options market. Apart from that, the deviations on both markets are small. The readings of deviation from put-call parity on the USD/INR options market are relatively sketchy as this market is often illiquid.

Realised volatility

Figure 5: Realised volatility

By 8th evening, realised vol on the Nifty futures market was showing some large values. Extreme realised volatility was found on the morning of the 9th: rvol was 6 times larger the pre-event mean value, as the market digested the two events. The price discovery was largely completed by 9th evening, and then realised vol was only slightly higher than the values seen in peacetime.

On the USD/INR futures, realised vol on the morning of the 9th was roughly 11 times larger than the normal values. The morning of 10th also shows a significant spike in realised vol.  While large price fluctuations were not evident on 10th, we observe a significant rise in variations on 11th.

Ruminating on methodology

Ordinarily, economists obsess on the question of identification. How do you know that event $x$ caused the outcome $y$? Could it be that there were other things going on which were impacting upon the observed change? We normally struggle to find plausible control units which can be juxtaposed against treatment units where both kinds of units are alike. The game is about rising beyond simple comparisons of means (or regressions), and look for plausible quasi-experimental designs. There are two tricks through which we are allowed to read the world and learn about how it works, without requiring the discipline of a matched sample of treatments and controls.

The first trick is when there are very big events. Ordinarily, we'd be worrying about whether the observed change in the treatment unit was caused by the event; what about other things that might be going on? But when big events happen, they dominate everything else. In the week under examination, we don't need to think about macroeconomic or firm news. The market was absorbed in doing price discovery in figuring out these two events; they drowned out all the other news flow.

The second trick is high frequency data. When we zoom into high frequency data, we have an opportunity to see the impact of the event alone, as it is unlikely that other confounding events have unfolded in that very short time.

Definitions

• Prices: These are traded prices of a security reported at minute-level frequency.
• Traded volumes: These are number of units of underlying security traded at five-minute frequency.
• No arbitrage violations: These are measured in terms of difference between the price of underlying security and price of near-month futures contract based on it. We do this for Nifty and for USD/INR but for gold, we just focus on the gap between MCX Gold futures and CME Gold futures.
• Violations in put-call parity: According to put-call parity, S + P = C + X(1+r)^-T i.e. a spot investment that is risk-managed using an at-the-money put option is tantamount to a combination of a bond and an at-the-money call option. Thus, violations in put-call parity are measured as the difference between the two investments.
• Realised volatility: It is constructed by computing intra-day returns at 5-second frequency. Thus, for every 5 minute-interval, realised volatility is computed as standard deviation across 60 readings of returns in the interval. We have very high frequency data and there are ample transactions within 5s, thus permitting differencing at such a high frequency.

Anurag Dutt, Sargam Jain and Susan Thomas are researchers at the IGIDR Finance Research Group. Ajay Shah is a researcher at the National Institute for Public Finance and Policy.

Methods for measurement of delays in the bankruptcy process

by Dhananjay Ghei and Shubho Roy.

When dealing with distress, the single big idea for avoiding value destruction is speed of action. Delays destroy value. When we cope with failing firms (through a bankruptcy code) or failing banks (through a resolution corporation), the measure of institutional quality is the recovery rate, and the key determinant of the recovery rate is speed of action.

Methods for measurement of delays

The conventional method for assessing delays is to run a survey of insolvency practitioners (judges, lawyers, accountants). In this article, we offer a new idea for measuring one part of the delay.

There are many parts of the bankruptcy process. Many societies try to avoid the problem of failed firms by putting things off. This leads to a delay between the actual firm failure and the commencement of the formal bankruptcy process. Could this be measured directly? We could start from the date of the end (the date of the official legal action about firm closure), and look back to the date on which extreme credit distress is identified. The gap between the two events will serve as a proxy for delays of the bankruptcy process.

Let's play this idea out with the failure of Kingfisher Airlines.

The decisive end date is 18th November, 2016, when the Karnataka High Court ordered the winding up of Kingfisher Airlines (See here).

Conventional notion of delay: The case was filed by an unsecured (foreign) creditor on 18th September, 2012. It took the legal system four years to come to the conclusion that Kingfisher Airlines is insolvent.

Looking back using interest coverage ratio (ICR) as a measure of credit stress. This is defined as the ratio of the earnings of a company (EBITDA) and the total interest payments due. The logic being that if a company's income is not enough to even pay the interest on its borrowing, there is acute distress. The literature in finance uses ICR in a couple of different ways when assessing firm distress. Here are three examples:

ICR below 1

Distress is where the ICR falls below 1 in any year. Example: Claessens, Djankov, and Lang 1998..

ICR below 0.75

Distress is where the ICR for any financial year falls below 0.75. Example: Love, 2010.

P-ICR

Distress is Persistent-ICR, which is defined as an ICR of less than 1 for three consecutive financial years. Example: Chung and Ratnovski, 2016

We extract the annual financial data for Kingfisher Airlines from CMIE Prowess database. The data is available from March 2001 till March 2013. Figure 1 below shows when Kingfisher Airlines met these criteria. It broke into ICR-below-1 and ICR-below-0.75 in March 2005 (red line) and it achieved P-ICR in March 2007 (orange line). The court finally ordered winding up 3520 days after Kingfisher meeting all the three criteria (gray line). The black line (October 20, 2012) is when DGCA suspended Kingfisher's license.

Figure 1: Interest cover ratio for Kingfisher Airlines

Looking back using Distance to Default as a measure of credit stress. Distance to Default (DtD) is measured as the difference between the asset value of the firm and the face value of its debt, scaled by the standard deviation of firm's asset value. It measures the distance (in standard deviations) between the expected value of the firm and the "default point" (face value of the debt). Thus, lower values of DtD imply that the firm is more likely to default on its financial obligations. DtD is implemented in R using the ifrogs package developed by IGIDR Finance Research Group.

Figure 2: Distance to default for Kingfisher Airlines

Figure 2 shows the DtD for Kingfisher Airlines from June 2007 onwards. As the listing took place on 12 June 2006, the DtD calculation only commences on 12 June 2007. The red line shows the date when DtD was below 1 for the first time (August 1, 2008). DtD reached its lowest value of 0.30 on July 6, 2009 (orange line). Thus, at that point the likelihood of default was the highest (19 percent). 

Conclusions

The approach shown above with Kingfisher is potentially scalable into large datasets and all countries. This could yield large-scale measurement, at the level of individual bankruptcy transactions, about delays. This can yield useful summary statistics.

The new insolvency law should reduce the time between financial insolvency and the legal recognition of the same. This should yield measurable gains when these techniques are applied to the data in the future.

Our approach yields micro data about delays which can be analysed in order to explore cross-sectional and time-series variation. Such analysis is not feasible with the conventional measurement of delays that is based on surveying practitioners.



Dhananjay Ghei and Shubho Roy are researchers at the National Institute for Public Finance and Policy. The authors would like to thank Radhika Pandey, Rajat Kochhar and Mohit Desai for their inputs.

Motivations for capital controls and their effectiveness

by Radhika Pandey, Gurnain K. Pasricha, Ila Patnaik, Ajay Shah.

The global financial crisis has re-opened the debate on the place of capital controls in the policy toolkit of emerging-market economies (EMEs). The volatility of capital flows during and after the global financial crisis, and the use of capital controls in major EMEs spawned a vigorous debate among policy-makers on the legitimacy and usefulness of capital controls.

In order to aid the development of best practices in capital controls policy, the literature needs to address four questions:

1. Under what circumstances do policy makers utilise capital controls? Do policy-makers use capital controls as macroprudential tools, as envisioned in the recent literature?
   
   
2. What impact do different capital controls have?
   
   
3. Do the benefits outweigh the costs?
   
   
4. How should real world institutional arrangements be constructed, to utilise these tools appropriately?
   
   

In a recent paper (Pandey et. al, 2016) we offer new evidence on the first and second of these questions.

A rich literature has sprung up in recent years, which has re-engaged with these questions. A number of recent studies examine effectiveness of controls in a single country (Brazil or Chile) or a multi-country setting. See for example, Alfaro et al, 2015; Fernandez et al., 2015; Forbes and Klein, 2015; Pasricha et al., 2015. A full list of references is in our paper. In this literature, several researchers have argued that capital controls may be particularly effective in a country like India with the legal and administrative machinery to implement controls (Habermeier et. al., 2011; Klein, 2012).

Indian policy makers have modified the capital control framework frequently to address concerns about the exchange rate, country risk perception and other issues. For example page 15 of RBI's 2014 Annual Report states that RBI's response to the developments following the US Fed's indication that it would taper its large-scale asset purchase program ``aimed at containing exchange rate volatility, compressing the current account deficit (CAD) and rebuilding buffers.'' This response included use of capital controls, foreign exchange intervention as well as interest rate changes. India is thus a good laboratory for studying the motivations and consequences of capital controls.

Credible research designs in this field require precise measurement of capital controls or capital control actions (CCAs). There are many concerns about the measurement obtained through conventional multi-country databases. We comprehensively analyse primary legal documents from 2004 to 2013, in order to construct a new instrument-level dataset about every capital control action for one asset class (foreign borrowing by firms) for one country (India).

In constructing this database, we differentiate between capital control announcements and capital control instruments (e.g., controls on minimum maturity of loans, controls on eligible borrowers, interest rate ceilings, etc.). In India, several instruments can be changed in the same announcement, and we count each instrument separately. We compare our approach with other recent work that compiles datasets on capital control actions (e.g. : Pasricha et al 2015; Pasricha 2012; Forbes et al. 2015) in our paper.

Q1: Under what circumstances do policy-makers utilise capital controls?

We use event studies to ask whether EME policy-makers use capital controls as macroprudential tools, as envisioned in the recent literature. Specifically, do EME policy-makers use capital controls to pursue macroprudential objectives or to achieve exchange rate objectives? A large literature since 2008 envisions capital controls as prudential tools, that can help mitigate systemic financial sector risk, and therefore views them in a more benign light than controls aimed at managing the exchange rate (See Korinek, 2011; Jeanne and Korinek, 2010; Bianchi, 2011, among others).

Factually assessing the motivations for past EME CCAs can help inform the debate on capital controls, as well as the resulting international consensus on the rules of governance for their use. On the one hand, if it can be discerned in the data that emerging markets have, in fact, been using capital controls to target systemic risk, this bolsters the legitimacy of the EME case for continued use of these instruments. On the other hand, if the data suggest that CCAs have been used for currency manipulation, this bolsters the case of those who argue that further international discussions on the rules of the game are needed to address multilateral concerns.

Figure 1: Exchange rate change prior to a easing CCA. Positive values denote depreciation.

Figure 2: Exchange rate change prior to a tightening CCA. Positive values denote depreciation.

The key result is in the two figures above. In the five weeks prior to an easing action, USD/INR depreciated by 3% on average. In the five weeks prior to a tightening action, USD/INR appreciated by 5% on average. Not only was the average trend prior to easing of inflow controls that of a depreciation of the currency, this also held true for the broad majority of events in sample: 42 out of the 68 instances of easing in our sample were preceded by exchange rate depreciation.

For the easing events which were preceded by an appreciation, the extent of the appreciation was small compared with that seen with events preceded by depreciation: the largest 5-week appreciation prior to an easing was 1.3%, compared to 9.2% for depreciation. The average appreciation prior to an easing was only 0.5%, compared to an average depreciation prior to easings of 5%.

None of the variables that measure the build-up of systemic risk show a similar strong pattern in the 6 months prior to the event date (see Figures 5-8 and Table 5 in the paper). The prime motivation for CCAs in India appears to be exchange rate policy and not macroprudential policy. This shows a certain gap between capital controls in the ideal world and capital controls as they operate in the field.

Q2: What impact do different capital controls have?

Next, we measure the impact of capital control actions. In order to obtain a credible estimation strategy, we utilise propensity score matching to identify time points which are counterfactual. This yields a quasi-experimental design where the treatment effect can be measured. Specifically, for each week in which a capital control action was taken, we identify a week in which macro / financial stress was similar, but no capital control action was taken.

Table 1: Causal impact of CCAs on various indicators

Impact upon	Coefficient	Std. Error	t-statistic
Credit growth	-0.44	1.7	-0.46
Stock prices	1.17	3.55	0.49
Frankel-Wei Residual	-0.23	0.92	-0.25
Net foreign inflow	-0.04	0.03	-1.33

Our results suggest that there was no significant impact of the capital control actions, either on the exchange rate or on measures connected with systemic risk (Table 1). Table 1 above shows the coefficient for the period 4 weeks after the capital control action. Similar values are found for all other time horizons. There is no statistically significant impact upon any of the outcomes at horizons from 1 to 4 weeks.

Broader implications of our results

These results have many implications for the global debate about capital controls. In many countries, the capital controls system was fully dismantled. In such an environment, it may be particularly easy to evade capital controls, for example through financial engineering. The best opportunity to obtain effectiveness of capital controls may be in countries like China or India, where large bureaucracies implement capital controls, and the detailed system of specifying rules about every asset class and every type of economic agent was never dismantled. For this reason, India is an ideal laboratory to study capital controls. If capital controls are found to be useful in India, the case could potentially be made that other EMEs, which dismantled the overall capital controls system, should reverse these reforms.

Our results show that Indian authorities seem to be using capital controls as a tool for exchange rate policy and not for systemic risk mitigation, and their actions seem to be ineffective. These results are also consistent with many papers in the recent literature which are skeptical about the usefulness of capital controls (Chamon and Garcia, 2015; Fernandez et. al, 2015; Forbes and Klein, 2015; Forbes et. al., 2015; Hutchison et. al, 2012; Klein, 2012; Patnaik and Shah, 2012; Pasricha et.al, 2015; Warnock, 2011).

The strength of the research presented here is that it provides credible estimates about one locale, India. A fruitful line of inquiry would be to apply such strategies to multiple countries, and build up a literature with careful assessment of country experience, one country at a time, about the ways in which capital controls are used, in the field, and about their treatment effects. A much more expansive strategy would seek to undertake such thorough instrument-level analysis on a multi-country scale in order to construct a consistent database about capital control actions on the scale of all EMEs or the whole world.

Even when capital controls do yield a desired treatment effect, the important question of cost-benefit analysis remains. A body of research is required which would assess the costs and the benefits of utilising these tools. On the cost-assessment side, a wide body of research on capital controls focuses on microeconomic distortions from capital controls (Alfaro et al, 2015; Forbes, 2007). On the benefits side, the evidence is mixed regarding the extent to which capital controls are able to deliver on the objectives of macroeconomic policy. While capital controls seem to be able to change the composition of flows toward more long-term debt, it is not clear to what extent this represents a mislabelling of flows (Magud et al., 2011; Carvalho and Garcia, 2008). Pasricha et al. (2015) find that capital control actions were not useful in allowing major emerging markets to change their trilemma configurations and Patnaik and Shah (2012) find that the Indian capital controls are not an effective tool for macroeconomic policy.

Further research is required on the institutional arrangements for capital controls. As an analogy, monetary policy was long viewed as being effective, but it was only in the 1980s that clarity was obtained around the institutional structure of independent central banks with inflation targets and monetary policy committees. In similar fashion, if capital controls have to graduate into the macroprudential policy toolkit, normative research is required in designing the optimal institutional arrangements for systemic risk regulation with mechanism design, akin to a monetary policy committee, and accountability, similar to an inflation target.

References

Laura Alfaro, Anusha Chari and Fabio Kanczuk. The real effects of capital controls: Financial constraints, exporters and firm investment NBER Working Paper 20726, Dec 2014.

Marcos Chamon and Marcio Garcia. Capital controls in Brazil: Effective? Journal of International Money and Finance, 2016 (Forthcoming).

Bernardo S. de M. Carvalho and Marcio G. P. Garcia. Ineffective controls on capital inflows under sophisticated financial markets: Brazil in the nineties In Sebastian Edwards and Marco G. P. Garcia (Eds.), Financial markets volatility and performance in emerging markets, pp. 29-96. University of Chicago Press.

Andres Fernandez, Alessandro Rebucci, and Martin Uribe. Are capital controls countercyclical? Journal of Monetary Economics, 76:1--14, 2015.

Anton Korinek. The new economics of capital controls imposed for prudential reasons. IMF Working Paper, Dec 2011.

Javier Bianchi. Overborrowing and systemic externalities in the business cycle. American Economic Review: Vol. 101 No. 7, Dec 2011.

Kristin J. Forbes. One cost of the Chilean capital controls: Increased financial constraints for smaller traded firms. Journal of International Economics 71(2): 294-323, Apr 2007.

Kristin J. Forbes and Michael W. Klein. Pick your poison: The choices and consequences of policy responses to crises. IMF Economic Review, 63(1):197--237, Apr 2015. ISSN 2041-4161.

Kristin J. Forbes, Marcel Fratzscher, and Roland Straub. Capital-flow management measures: What are they good for? Journal of International Economics, 96, Supplement 1:S76 -- S97, 2015. ISSN 0022-1996. 37th Annual NBER International Seminar on Macroeconomics.

K. F. Habermeier, C. Baba, and A. Kokenyne. The effectiveness of capital controls and prudential policies in managing large inflows. IMF Staff Discussion Note SDN/11/14, International Monetary Fund, 2011.

Nicolas E. Magud, Carmen M. Reinhart and Kenneth S. Rogoff. Capital controls: Myth and reality - A portfolio balance approach. NBER Working Paper No. 16805, Feb, 2011

Michael M. Hutchison, Gurnain Kaur Pasricha, and Nirvikar Singh. Effectiveness of capital controls in India: Evidence from the offshore NDF market. IMF Economic Review, 60(3): 395--438, 2012.

Michael W. Klein. Capital controls: Gates versus walls. Brookings Papers on Economic Activity, 45(2 (Fall)):317--367, 2012.

Olivier Jeanne and Anton Korinek. Excessive Volatility in Capital Flows: A Pigouvian Taxation Approach. American Economic Review, 100(2), May 2010.

Radhika Pandey, Gurnain Kaur Pasricha, Ila Patnaik, Ajay Shah. Motivations for capital controls and their effectiveness. Working paper, 2016.

Gurnain Kaur Pasricha. Recent trends in measures to manage capital flows in emerging economies. The North American Journal of Economics and Finance 23 (3), 286-309.

Gurnain Kaur Pasricha, Matteo Falagiarda, Martin Bijsterbosch and Joshua Aizenman. Domestic and multilateral effects of capital controls in emerging markets. NBER Working Paper No. 20822.

Ila Patnaik and Ajay Shah. Did the Indian capital controls work as a tool of macroeconomic policy. IMF Economic Review, 60(3):439--464, 2012.

Frank E. Warnock. Doubts about capital controls. Working Paper 14, Council on Foreign Relations, 2011.




Gurnain Pasricha is at the Bank of Canada, and the other three authors are at the National Institute for Public Finance and Policy, New Delhi. The views expressed in this post are those of the authors. No responsibility for them should be attributed to the Bank of Canada or NIPFP.

Looking beyond the label `algorithmic trading'

by Ajay Shah.

At the EMF 2015 conference, I attended a talk by Pradeep Yadav, where he presented a paper: Raman, Robe, Yadav, 2015. This paper analyses algorithmic traders ("AT") and manual traders ("MT") on one of the world's largest electronic limit order book exchanges, the National Stock Exchange ("NSE").

NSE is an ideal laboratory for studying these questions, as it is a simple plain limit order book market, without the confusion caused by market makers. In addition, the overall equity market structure is simple, with two exchanges (NSE and BSE) where NSE has dominant market share. The complexities of the fragmented order flow and multiple trading venues, of the US, is not present. At NSE, both spot and futures trade in the same time zone, with orders emanating from the same co-location facility, which also facilitates research. NSE data has thus shaped up to be a very nice foundation for understanding markets, in recent years, with some of the cleanest microstructure work getting done here.

The puzzle

Pradeep Yadav and his coauthors find that under conditions of market stress, ATs withdraw from the market. This immediately lends itself to a pejorative interpretation: "ATs are good for liquidity under good times, but are quick to withdraw when the going gets difficult, and this is creating a new set of problems".

I wondered how this squares with a powerful result from Aggarwal & Thomas, 2014. This is a modern causal econometrics paper, and in this they find that the incidence of mini flash crashes goes down when there is more AT. They look for mini flash crashes defined as 2%, 5% and 10% declines of the price within a five minute window (see Table 6, page 28). All three coefficients are negative; more AT gives fewer flash crashes. For the 2% and 10% case, the differences are not statistically significant, but for the case of a 5% drop of prices in 5 minutes, bigger AT gives a statistically and economically significant decline in the incidence of flash crashes.

Both papers seem to have persuasive empirical strategies. How do we square the results? How is it that ATs are more likely to step away in difficult times (Raman, Robe, Yadav) but at the same time how it is that when there is more AT, mini flash crashes are less frequent (Aggarwal, Thomas)?

A better classification system

In order to figure out what's going on, I think we should break with the classification AT vs. MT. Instead, it's better to think in terms of simple, mechanistic trading strategies vs. complex strategies that involve human judgment. For the purpose of argument, let's call these "SI" for simple vs. "CO" for complex strategies.

Let's start with the old world, before algorithmic trading. In that world, we very much had many humans running SI strategies and many humans running CO strategies. `Technical analysis', and other trend following mechanistic strategies, were around well before algorithmic trading came along!

As Friedman, 1953, reminded us, there is a Darwinian process at work where speculators who lose money tend to exit the market. Because markets are competitive, the dumb adherence to a SI strategy would induce losses, and the people who did this would exit the market. Hence, the only sensible approach for a trader who uses a SI strategy is to either stop some times (i.e. have a "kill switch") or switch to a CO strategy at certain times.

In the good old days, SI speculators had "kill switches". When the market got weird, they would just stop trading. Nobody expected a simple trend following speculator to behave unchanged when volatility changed or when big news broke. We had trading floors where a boss would shut down some strategies from time to time. This was akin to a "kill switch" applied to a large number of SI strategies.

All that has happened with algorithmic trading is that we now have powerful clerks, i.e. computers, who are the foot soldiers implementing SI strategies. Nothing else has changed. Humans are still in charge!

Some human traders keep a swarm of SI strategies under leash, and when market conditions get difficult, they hit the kill switch. Some of them switch to CO strategies when the going gets difficult, and because CO strategies are much harder to program, they may well do this trading by hand. Some bosses of trading floors yank hundreds of SI strategies when the going gets weird.

Resolving the puzzle

There were always SI strategies and CO strategies. These have been around ever since organised financial trading began. In the older data, we are not able to disentangle the two.

In recent years, the SI strategies have gotten automated. We have reduced the use of humans in mechanistic tasks, and got computers to do this clerical work. For the first time in human history, we are now seeing a flag on orders where orders from SI strategies are now called "AT" orders. The CO strategies continue to be mostly done by hand, as it's quite hard doing this programming.

There is nothing wrong or unusual in SI strategies backing out of the market when conditions become confusing. SI strategies can only work in peaceful times. A trader who ran SI strategies all the time would exit the market as his wealth would run out (Friedman, 1953).

SI strategies done through AT give us more eyeballs looking at the millions of traded products in the modern exchange environment. When there's a dislocation in a market (e.g. a crash in the futures price), immediately, hundreds of traders come through with a mechanistic response (reverse cash and carry arbitrage), which stabilises the price. In contrast, in the manual world, the field of view of each human was limited, and when a little crash got started, there were fewer people available to interfere with it. This gave more mini flash crashes in the pre-AT world. This is how both statements are correct:

1. In times of market stress, the AT orders shy away (Raman, Robe, Yadav, 2015)
2. Greater AT intensity reduces the incidence of mini flash crashes (Aggarwal & Thomas, 2014).

References

Nidhi Aggarwal, Susan Thomas. The causal impact of algorithmic trading on market quality, Working Paper, 2014.

Milton Friedman. The case for flexible exchange rates. In Essays in Positive Economics, The University of Chicago Press, 1953.

Vikas Raman, Michel A. Robe, Pradeep K. Yadav. Man vs. Machine: Liquidity Provision and Market Fragility, Working Paper, 2015.

The changing landscape of equity markets

by Nidhi Aggarwal and Chirag Anand.

The arrest of a London based algorithmic trader, Navinder Singh Sarao, on charges of triggering the US flash crash of 2010 has once again brought regulatory concerns on high frequency trading (HFT) to the forefront. With the underlying fear that the use of high speed complex algorithms can pose systemic risk, regulators worldwide are considering actions to tighten their grip on HFT. The Indian securities markets have not remained immune to such concerns, and the securities market regulator, SEBI, has indicated that steps will be taken to keep the level of algorithmic trading (AT) in check. Very recently, even RBI in its annual Financial Stability Report expressed its concerns regarding high levels of algorithmic orders in the Indian securities market.

Despite all the fears and the measures that are being taken to curb HFT, one needs to note that the evidence regarding how HFT (or AT) hurts the market is yet to be established. Concerns such as higher percentage of algorithmic orders creates higher level of systemic risk in the financial system are not backed by strong empirical evidence. Studies examining AT/HFT trading only find evidence contrary to this popular notion (Brogaard et al., 2015; Thomas and Aggarwal, 2014). Other studies (Biais and Faoucault, 2014) examining the overall effect of AT/HFT on market quality find that higher levels of AT/HFT improves market quality by increasing liquidity and price efficiency. In spite of this overwhelming evidence on the effect of AT, regulatory fears on how increased market complexity can disrupt the financial markets remain.

An analysis at the Finance Research Group, IGIDR aims to provide a few insights on the proliferation of HFT (or AT) in the Indian markets. Using a unique tick by tick orders and trades dataset from one of the most liquid stock exchanges in the country, the National Stock Exchange (NSE), we examine how AT/HFT has changed the equity market structure in India. In addition to the usual details of price and volume, the data contain details of whether an order was sent by an AT or a non AT, and whether the order was a new order, or an old order that was modified or cancelled. A clear demarcation of orders sent by AT versus non AT, enables us to examine the characteristics of how AT's trade in the markets vis-a-vis non AT.

We analyse two periods: a low AT period (November-December 2009) and a high AT period (November-December 2013). Few points emerge:

• Between the two periods, percentage of orders entered by algorithmic traders increased from 11.36% to 62.76% on equity spot, from 38.93% to 93.72% on single stock futures (SSF), and from 21.29% to 86.79% on single stock options (SSO).
• On the most liquid segment of NSE, that is the Nifty options, the percentage of new orders entered by AT increased from 19.60% to 93.56%.
• On Nifty futures contract, it increased from 21.57% to 91.23%.

The values indicate that a large proportion of the orders that are entered on NSE today are by algorithmic traders (AT). A majority of these orders are limit orders, indicating that instead of going for the special orders that the exchange offers, AT prefer the traditional limit orders which offer them greater flexibility to manage their orders.

Do AT supply liquidity or demand liquidity?

The increase in percentage of AT orders in the market raises the concern on whether that increase corresponds to a similar increase in liquidity supply, or, whether they consume liquidity from non algorithmic traders. For each segment on NSE, we analyse the percentage of trades in which AT supplied liquidity versus the trades where they demanded liquidity. When an order that comes to the market trades against an existing order in the book, the new order is said to have taken (demanded) liquidity, while the existing order is said to have provided (supplied) liquidity.

The graph above indicates the share of AT orders in total liquidity demanded increased across all the segments between the two periods. However, this matches with their share of orders in total liquidity supplied to the market in all except the Nifty options market. We further break this analysis into who supplies liquidity to whom. This is depicted in the following graph.

In the above graph, the top-left panel indicates the percentage of trades in which AT demanded liquidity from another AT. On the spot market, for example, AT took liquidity from other AT in 6.34% of trades in 2013. The top-right panel indicates the percentage of trades in which non AT demanded liquidity from AT. The bottom left panel indicates the percentage of trades in which AT demanded liquidity from non AT. Finally, the bottom right panel indicates the percentage of trades in which non AT demanded liquidity from non AT.

A difference in the values in the bottom right panel from 100 indicates the AT-intensity, that is the percentage of trades that occurred on NSE in which AT was either on one or both sides of the trade. For example, on the spot market, in 2013, the percentage of trades in which AT were present atleast on one side of the trade was (100 - 44.3)% = 55.7%.

Of particular interest are the top-right and bottom-left graphs. These two graphs indicate non AT demand for liquidity from AT, and AT demand for liquidity from non AT, respectively. The values in the graph reinforce the observation that AT demand as much liquidity from non AT as they supply to them for all except the Nifty options market.¹ This suggests that the concern that AT consume liquidity from non AT does not hold.

We now proceed on to examining how the order placement strategies of AT have changed the market structure on NSE.

Changing market structure due to high speed access

Q:1 How have order placement strategies changed after faster market access? With a majority of the orders coming from AT, we first examine if there has been a change in the order placement strategies by market participants. Specifically, we examine if the increase in the number of orders has translated into a larger number of trades, or are most of the orders that are entered are eventually cancelled?

The table below indicates the percentage of orders that get traded and cancelled by AT and non AT.

All values as % of total orders entered
	Spot		SSF		SSO		Nifty futures		Nifty options
	2009	2013	2009	2013	2009	2013	2009	2013	2009	2013
AT	12.42	62.19	39.18	93.30	20.56	84.89	11.11	87.84	21.71	93.38
Traded	3.91	12.37	1.59	2.20	0.74	2.61	3.02	7.73	1.49	7.47
Cancelled	8.31	49.73	37.52	90.91	19.65	82.03	7.99	79.88	20.15	85.88
Non AT	87.58	37.81	60.82	7.70	79.44	15.11	88.89	12.16	78.29	6.62
Traded	56.11	25.69	14.17	3.00	24.95	6.03	45.37	8.18	32.76	4.43
Cancelled	21.75	7.24	44.88	3.20	44.05	6.52	39.67	2.70	43.22	1.63

The first row in the table indicates the percentage of orders entered by AT. As discussed eariler, the share of AT in the total number of orders sent to the NSE has risen significantly. The second row in the table indicates the percentage of AT orders that got traded.

The table shows that the increase in percentage of new orders entered by AT is not matched with a higher percentage of orders that got traded. Instead, we see a decline in the percentage of traded orders across all the five segments (spot, SSF, SSO, Nifty futures, Nifty options). For example, on the spot market, the percentage of orders that got traded declined from 60.02% in 2009 to 38.06% in 2013. We also find a significant increase in the percentage of orders that got cancelled in the high AT period (2013). Of the total unique orders that came to NSE, the percentage of orders that got cancelled increased from 30.06% in 2009 to 56.97% in 2013 on the spot segment, from 82.40% to 94.11% on the SSF and from 63.70% to 88.55% on the SSO. On Nifty futures, this percentage increased from 47.66% to 81.58% and on Nifty options from 63.37% to 87.51%.

While there could be legitimate reasons for such cancellations (Hasbrouk and Saar, 2009), the increase in the percentage of cancelled orders raises concerns about phantom liquidity (also known as spoofing, flickering quotes, or fleeting liquidity), that is, the fear that high speed access allows the trader to post an order for everyone to see, but withdraws it before anyone can act on it. We examine this concern in the next question.

Q:2 Do order cancellations occur at very short intervals? Higher percentage of order cancellations, by itself is not a matter of concern. The concern instead is that these orders might be getting cancelled in such short a time that other traders, who do not have the advantage of fast market access, are unable to execute their orders against such orders. Or, these orders could be sending signals of false liquidity. In order to pin down these concerns, the evidence of cancellations needs to be combined with evidence of speed of cancellations - or the lifespan of the orders. If a majority of the orders are cancelled in very short time intervals, then it could be suggestive of phantom liquidity in the markets.

Cancelled orders as a percentage of total orders entered on Nifty options

Cancelled orders as a percentage of total orders entered on SSF

The graphs above indicate the percentage of orders that got cancelled in less than a second on the two most liquid NSE segments: Nifty options and single stock futures (SSF). The graphs suggests that in the high AT period (2013), more than 70% of the orders entered on the SSF and about 54% of the orders entered on the Nifty options market got cancelled within a second.² These values are substantially higher than the values in the low AT period of 2009, during which 7.83% and 14.96% of orders got cancelled within one second on SSF and Nifty options.

A useful question to ask is how these numbers compare with the global markets. A similar analysis for the US equity markets by SEC indicates that 45.9% of the orders were cancelled within a second during Q2 2013.³
 
Q:3 Is fast too fast? The analysis above indicates that high speed access has made cancellations too fast. The next question that becomes important to ask is, ``Is this too fast''? To characterise the intensity of what is fast, we use the SEC's approach. In a speech in April 2014, by the then Associate Director of SEC, Gregg Berman, noted:

``If the speed of cancellation is much quicker than the speed at which those quotes can be accessed, then I would say quote cancellations are not only fast, but perhaps they are too fast. However, if market participants can lift quotes just as quickly as others can cancel them, I would say that the cancellations might be fast, but not necessarily too fast."

And its relevance in informing the policy debate:

``If quote cancellations are indeed too fast for the rest of the market to keep up, it might make sense to slow down this particular aspect of the markets, perhaps with some sort of minimum quote-life requirement. But it the data shows that at least some market participants can access quotes just as quickly as they can be canceled, this suggest that both sides of the market are very fast and if you want to slow down the market -- in a way that does not bias one side, you would need to not only address the speed of quote cancellations, but also the speed at which liquidity is taken."

We examine this by comparing the lifespan of cancelled orders with that of the traded orders. We once again restrict our discussion to the two most liquid segments on NSE: Nifty options and SSF.

The graph above shows the results for Nifty options for cancelled (top-panel) and traded (bottom-panel) orders. A shift from the red to yellow region indicates increase in the speed of order cancellations or execution. In 2009, while about 30% of all cancelled orders remained in the book for less than a second, about 55% of all traded orders were the result of some trader hitting limit orders within that same time period. These numbers rose to 65% and 80% respectively in 2013. Also noticeable is that the number of modifications on these cancelled orders is in the range of 0-5. This suggests two features of trading activity on the Nifty options market:

1. A majority of the orders that get cancelled do not undergo large number of modifications.
2. Access to speed has indeed increased the speed of order cancellations, but this speed is lower than the speed of execution.

The Nifty options inferences do not however hold for the SSF. In 2009, the percentage of cancelled orders within a lifespan of less than a second on SSF was almost negligible, while the percentage of traded orders within the same lifespan was less than 40%. The numbers changed dramatically in 2013. The graph shows a shift from red to yellow region for cancelled orders, but only a shift from red to orange region for traded orders. The percentage of cancelled orders with a lifespan of less than a second was about 75%, while the percentage of traded orders within the same lifespan was about 45%. This indicates that the speed of order cancellations surpassed the speed of trade executions in 2013.

Summary

In a nutshell, the findings can be summarised as:

1. The share of algorithmic orders in total orders that come to the market has risen significantly.
2. Except for the Nifty options market, the share of algorithmic traders in liquidity demand matches with their share in liquidity supply.
3. A large majority of the orders on NSE are cancelled, with most of them occurring within a second of order entry.
4. The speed of order execution is higher than the speed of order cancellations on Nifty options. This is however not true of the SSF segment of the NSE.

The above analysis does imply that the order placement activities have changed significantly with a lot of cancellations occurring within short time-frames. However, to analyse whether this degree of cancellations could be hurting the other market participants, it is critical to examine the where these quote cancellations occur? If most of these cancellations are occurring around the best bid and ask prices (or even the upto level 5 depth of the market), such cancellations could be a cause of concern. Further analysis aims to capture this aspect.

References:

 
High-frequency trading and extreme price movements by Brogaard J, Carrion A, Moyaert T, Riordan R, Shkiklo A and Sokolov K, 2015, Working Paper

The causal impact of algorithmic trading on market quality by Susan Thomas and Nidhi Aggarwal, 2014. IGIDR Working Paper.

HFT and market quality by Biais B and Foucault T, 128, 2014 in Bankers, Markets and Investors, p. 5-19.

Technology and liquidity provision: The blurring of traditional definitions by Hasbrouck J, Saar G, 2009. Journal of Financial Markets, Volume 12, Issue 2, May 2009, p. 143-172.

---

 

Footnotes

1. The reason for the difference in the nature of AT liquidity demand and supply on the options market needs further investigation.
2. We record similar values for the rest of the market.
3. The findings are also comparable to studies investigating fleeting orders. For example, Hasbrouck and Saar (2009) find that 36.69% of the limit orders get cancelled in less than two seconds on INET.