Short squeeze by the numbers
16 February 2021
Sam Pierson, IHS Markit securities finance data analyst, breaks down the uses and limitations of exchange short interest data and explains how the SI Forecast solution aims to do better
Image: Shutterstock
A well-known use for US equity finance data is estimation of the short interest (SI) published by US exchanges in between the bi-monthly publications of that dataset. This can be done effectively and result in a timeliness and level of insight unavailable from the exchange SI data alone. There are important considerations when applying these methods, which we’ll discuss here. These considerations are critical for market participants, highlighted by the January short squeeze when a demand for real-time insights faced the real-world challenge of providing them. Estimating exchange SI with a model has advantages and drawbacks which are revealed by reviewing the model output in the context of the inputs.
The bi-monthly exchange SI publications show the gross short positions across underlying accounts held with FINRA member broker-dealers. The disclosures are aggregated and made available after the close on the seventh trading following the settlement date they were collected for. The 29 January SI dataset was published by US exchanges after the close on 9 February. Since the dataset was collected for 29 January settlement, it reflects settled open short positions held from trade-date 27 January.
Equity finance data including the number of shares on loan, is also published for a settlement date and reflects the trade date two days prior. This brings in an essential consideration for estimating the SI: the number of shares on loan, as published by IHS Markit Securities Finance, can be considered the net borrow demand beyond broker-dealers’ internal supply of shares.
When a short sale is made, the broker settling the trade can either borrow shares from an external counterparty (picked up in the equity finance dataset) or they can use shares already in their custody for delivery to the counterparty who made the purchase from the short seller. For prime brokers (PB), who handle most short sales, the key sources of internal supply are hedge fund longs (which may be in margin or fully paid accounts) and delta-one long positions. The gap between the SI and shares on loan can generally be interpreted as the internal supply the brokers held in custody and delivered to settle short sales.
The management of dynamic internal supply means that brokers will modify their borrowing both to reflect changes in client short positions and to reflect changes in their own internal availability. They may also modify their usage of internal supply in anticipation of supply or demand changes, which is why shares on loan may react to an event on trade date or T+1. This creates a challenge for estimating the SI in real-time based on shares on loan, because it’s possible that a decrease in borrowing reflects an increase in internal availability as opposed to a decrease in short positioning.
To deal with share borrowing changes potentially being driven by variations in broker-dealer longs, a model which estimates the SI needs to qualify the changes in shares on loan by estimating the probability that they reflect a change in SI. The SI Forecast from IHS Markit does this partly by looking at the historical relationship between the SI and shares on loan. The goal is to correctly forecast changes with a minimum of error introduced by forecasting large changes that don’t materialise. For that reason, the SI Forecast incorporates the most recently published exchange SI data and subsequently adjusts based on model inputs including shares on loan. The forecast is unlikely to fully reflect a change in shares on loan unless the two series are very similar historically, which suggests a minimal PB internal availability.
GameStop example:
The 15 January NYSE SI dataset showed 61 million shares short for GameStop, a 9.4 million share decline since the 31 December publish. The 15 January dataset was published after the close on 27 January. The timeline is important to note because during the trading sessions from 22-27 January, when the share price increased from $65 to $347, the most recently available NYSE SI data was the 31 December observation, which showed 71 million shares short.
GameStop shares on loan was published by IHS Markit as 51 million shares for 15 January, 9.4 million shares lower than the NYSE SI for 15 Jan settlement. The 15 January equity finance dataset was published on 18 January (S+1, with a weekend). By 27 January, when the 15 January SI was published, the most recently published equity finance dataset was 26 January, which showed 41.7 million shares on loan, a 9.3 million share reduction since 15 January.
From the perspective of post-close 27 January, market participants knew that the SI for 15 January was 61.8 million shares, which declined by 9.4 million shares since 31 December. The 26 January shares on loan showed a decline of 9.3 million shares as compared with 15 January; if that were fully reflected in a reduction in SI, that would mean 52.5 million shares short on 27 January.
The first SI forecast published by IHS Markit which reflected knowledge of 15 January SI was for the 27 January dataset. That was published on 28 January and estimated the SI at 56.4 million shares. The forecast was heading in the direction of assuming the change in shares on loan was the change in SI (52.5 million shares). However, the model gave a lower weighting to the equity finance data and was nearly 4 million shares higher for that reason.
Over the final days of January, the shares on loan continued to decline. From the perspective of 1 February, when the 29 January equity finance dataset had been published, there were only 17.4 million shares on loan, reflecting a 31.5 million share reduction in shares on loan between 15-29 January. The most recent SI was still 15 January at 61.7 million shares, so if the change in shares on loan were fully reflected, that would mean 30 million shares short. The SI Forecast published on 1 February estimated 50 million shares short.
On 9 February, the SI dataset for 29 January settlement was published, which showed that SI had declined by 40 million shares to 21.4 million shares. The gap between the shares on loan and SI declined from 10.7 million shares on 15 January to 1.9 million shares on 29 January. The declining gap indicated the possibility that hedge fund longs who had previously lent their shares (in so doing reducing the need for their brokers to borrow shares externally for client shorts) had recalled their shares over the last two weeks of January, forcing a larger portion of the total short position to be settled with shares borrowed from the equity finance channel. The 29 January settlement pertains to 27 January trade date, which means the vast majority of the short position had been covered by trade date 27 January.
The question may then be asked: why bother with the forecast? The purpose of the forecast is to provide a daily estimate of the exchange SI which will be as close as possible to the as-yet unpublished SI figure. The previous publication of that figure will, in general, be reasonably close, so the assumption of no-change will yield a result that appears accurate in comparison to a specific date, but obviously does nothing to track changes between publishes. There is a persistent correlation between the shares on loan and SI for many US equities, so using the shares on loan in a model makes sense; however, there are known causes for the series to diverge (changes in broker-dealer internal availability). Any suggestion of change from the prior SI has the potential to introduce error, so a substantial recognition of changes in shares on loan should only be done when the two series are highly correlated, grading slowly toward a very limited reliance on equity finance data where there is a low expectation for forecasting success. In this view, the forecast performed as expected with the inputs available. It would have been possible for the 29 January SI to print at 50 million shares, which would have been interpreted as a substantial uptick in dealer inventory, likely the result of an increase in hedge fund longs (possibly also some index related delta-one longs). Given the events which unfolded over the last week of January, along with the decline in shares on loan, that may have been deemed unlikely, but is important not to discount as a possibility when considering the model output.
The exchange SI is a valuable source of data which is widely used. When the phrase ‘short interest’ is used pertaining to US equities it is understood to mean the bi-monthly figures. One important caveat is that exchange SI only includes short positions cleared by FINRA member broker-dealer entities, which means it is possible short positions will not to be included.
While the knowledge that the SI data is not comprehensive is important, the narrative impact from the common knowledge of SI may be more significant than ever. The reporting lag and the lag to trade date mean that the shortest time between a short sale being traded and included in the SI dataset is 10 trading days. On the eve of an SI release, the prior publish is more than three weeks old. Equity finance data can help identify short flows between publishings, and the comparison reveals other signals; however, it is no panacea. Using a model to estimate SI can be helpful but introduces a new source of error and should therefore, when precision is required, be considered in the context of model inputs.
The bi-monthly exchange SI publications show the gross short positions across underlying accounts held with FINRA member broker-dealers. The disclosures are aggregated and made available after the close on the seventh trading following the settlement date they were collected for. The 29 January SI dataset was published by US exchanges after the close on 9 February. Since the dataset was collected for 29 January settlement, it reflects settled open short positions held from trade-date 27 January.
Equity finance data including the number of shares on loan, is also published for a settlement date and reflects the trade date two days prior. This brings in an essential consideration for estimating the SI: the number of shares on loan, as published by IHS Markit Securities Finance, can be considered the net borrow demand beyond broker-dealers’ internal supply of shares.
When a short sale is made, the broker settling the trade can either borrow shares from an external counterparty (picked up in the equity finance dataset) or they can use shares already in their custody for delivery to the counterparty who made the purchase from the short seller. For prime brokers (PB), who handle most short sales, the key sources of internal supply are hedge fund longs (which may be in margin or fully paid accounts) and delta-one long positions. The gap between the SI and shares on loan can generally be interpreted as the internal supply the brokers held in custody and delivered to settle short sales.
The management of dynamic internal supply means that brokers will modify their borrowing both to reflect changes in client short positions and to reflect changes in their own internal availability. They may also modify their usage of internal supply in anticipation of supply or demand changes, which is why shares on loan may react to an event on trade date or T+1. This creates a challenge for estimating the SI in real-time based on shares on loan, because it’s possible that a decrease in borrowing reflects an increase in internal availability as opposed to a decrease in short positioning.
To deal with share borrowing changes potentially being driven by variations in broker-dealer longs, a model which estimates the SI needs to qualify the changes in shares on loan by estimating the probability that they reflect a change in SI. The SI Forecast from IHS Markit does this partly by looking at the historical relationship between the SI and shares on loan. The goal is to correctly forecast changes with a minimum of error introduced by forecasting large changes that don’t materialise. For that reason, the SI Forecast incorporates the most recently published exchange SI data and subsequently adjusts based on model inputs including shares on loan. The forecast is unlikely to fully reflect a change in shares on loan unless the two series are very similar historically, which suggests a minimal PB internal availability.
GameStop example:
The 15 January NYSE SI dataset showed 61 million shares short for GameStop, a 9.4 million share decline since the 31 December publish. The 15 January dataset was published after the close on 27 January. The timeline is important to note because during the trading sessions from 22-27 January, when the share price increased from $65 to $347, the most recently available NYSE SI data was the 31 December observation, which showed 71 million shares short.
GameStop shares on loan was published by IHS Markit as 51 million shares for 15 January, 9.4 million shares lower than the NYSE SI for 15 Jan settlement. The 15 January equity finance dataset was published on 18 January (S+1, with a weekend). By 27 January, when the 15 January SI was published, the most recently published equity finance dataset was 26 January, which showed 41.7 million shares on loan, a 9.3 million share reduction since 15 January.
From the perspective of post-close 27 January, market participants knew that the SI for 15 January was 61.8 million shares, which declined by 9.4 million shares since 31 December. The 26 January shares on loan showed a decline of 9.3 million shares as compared with 15 January; if that were fully reflected in a reduction in SI, that would mean 52.5 million shares short on 27 January.
The first SI forecast published by IHS Markit which reflected knowledge of 15 January SI was for the 27 January dataset. That was published on 28 January and estimated the SI at 56.4 million shares. The forecast was heading in the direction of assuming the change in shares on loan was the change in SI (52.5 million shares). However, the model gave a lower weighting to the equity finance data and was nearly 4 million shares higher for that reason.
Over the final days of January, the shares on loan continued to decline. From the perspective of 1 February, when the 29 January equity finance dataset had been published, there were only 17.4 million shares on loan, reflecting a 31.5 million share reduction in shares on loan between 15-29 January. The most recent SI was still 15 January at 61.7 million shares, so if the change in shares on loan were fully reflected, that would mean 30 million shares short. The SI Forecast published on 1 February estimated 50 million shares short.
On 9 February, the SI dataset for 29 January settlement was published, which showed that SI had declined by 40 million shares to 21.4 million shares. The gap between the shares on loan and SI declined from 10.7 million shares on 15 January to 1.9 million shares on 29 January. The declining gap indicated the possibility that hedge fund longs who had previously lent their shares (in so doing reducing the need for their brokers to borrow shares externally for client shorts) had recalled their shares over the last two weeks of January, forcing a larger portion of the total short position to be settled with shares borrowed from the equity finance channel. The 29 January settlement pertains to 27 January trade date, which means the vast majority of the short position had been covered by trade date 27 January.
The question may then be asked: why bother with the forecast? The purpose of the forecast is to provide a daily estimate of the exchange SI which will be as close as possible to the as-yet unpublished SI figure. The previous publication of that figure will, in general, be reasonably close, so the assumption of no-change will yield a result that appears accurate in comparison to a specific date, but obviously does nothing to track changes between publishes. There is a persistent correlation between the shares on loan and SI for many US equities, so using the shares on loan in a model makes sense; however, there are known causes for the series to diverge (changes in broker-dealer internal availability). Any suggestion of change from the prior SI has the potential to introduce error, so a substantial recognition of changes in shares on loan should only be done when the two series are highly correlated, grading slowly toward a very limited reliance on equity finance data where there is a low expectation for forecasting success. In this view, the forecast performed as expected with the inputs available. It would have been possible for the 29 January SI to print at 50 million shares, which would have been interpreted as a substantial uptick in dealer inventory, likely the result of an increase in hedge fund longs (possibly also some index related delta-one longs). Given the events which unfolded over the last week of January, along with the decline in shares on loan, that may have been deemed unlikely, but is important not to discount as a possibility when considering the model output.
The exchange SI is a valuable source of data which is widely used. When the phrase ‘short interest’ is used pertaining to US equities it is understood to mean the bi-monthly figures. One important caveat is that exchange SI only includes short positions cleared by FINRA member broker-dealer entities, which means it is possible short positions will not to be included.
While the knowledge that the SI data is not comprehensive is important, the narrative impact from the common knowledge of SI may be more significant than ever. The reporting lag and the lag to trade date mean that the shortest time between a short sale being traded and included in the SI dataset is 10 trading days. On the eve of an SI release, the prior publish is more than three weeks old. Equity finance data can help identify short flows between publishings, and the comparison reveals other signals; however, it is no panacea. Using a model to estimate SI can be helpful but introduces a new source of error and should therefore, when precision is required, be considered in the context of model inputs.
NO FEE, NO RISK
100% ON RETURNS If you invest in only one securities finance news source this year, make sure it is your free subscription to Securities Finance Times
100% ON RETURNS If you invest in only one securities finance news source this year, make sure it is your free subscription to Securities Finance Times
