Robert Almgren (Quantitative Brokers): Extended Microprice for Price Interpolation
It is well-known that order book information beyond the inside quotes can provide meaningful information about the fair value of a traded asset, and useful predictions for future values. There are many ways in which a generalized price interpolant can be constructed. We outline the necessary properties that such an interpolant should have, we present one implementation, and we give empirical evidence for its predictive ability.
Hanna Assayag (HSBC): TBD
Markus Baldauf (UBC): Competition and Privacy in Off-Market Trading
In the U.S., brokers route most retail equity orders to wholesalers, who execute them off-exchange at better prices than the NBBO. We propose a new theory for the willingness of wholesalers to offer such off-exchange discounts. Our theory not only is consistent with industry anecdotes, but also explains otherwise puzzling empirical patterns, and has important implications for proposed reforms. In our model, each wholesaler privately learns from off-exchange retail orders she executes, which provides her with information rents in subsequent on-exchange market making. Wholesalers compete for retail orders with favorable prices in order to obtain these information rents. This mechanism can predict midpoint or better off-exchange prices, a commonly observed pattern that classic adverse selection or inventory models have difficulty explaining. In light of our mechanism, we then compare the status quo with three alternative policies for off-exchange trading: (1) enhanced trade reporting merely eliminates information rents, worsening retail prices; (2) order-by-order auctions may improve prices by enhancing wholesaler competition, but their public nature may also harm prices by eliminating information rents. We propose (3) a request for stream (RFS) design which both increases competition and preserves information rents, unambiguously improving retail prices.
Neil Chriss (Paloma Partners): TBD
Robert Graumans (AFM and University of Oxford): Anonymity, Signaling, and Collusion in Limit Order Books
Anonymity is a fundamental design feature of many limit order books. Yet, by analyzing data that includes trader identifiers, we show that market makers actively undermine this anonymity. They do so by placing large-volume limit orders, effectively signaling their presence to other market makers. In addition, they hardly trade with one another and strategically target— or "snipe"—retail limit orders.
To explain this behavior, we develop a model that incorporates both competitive and collusive market equilibria. The model shows that market makers’ actions align with a collusive equilibrium, where signaling is used to prevent mutual sniping. This signaling mechanism allows market makers to coordinate and share the order flow from retail traders, while simultaneously suppressing competition from those same retail orders. As a result, market makers attract additional benign flow from impatient investors who might otherwise have executed trades against retail limit orders.
Björn Hagströmer (Stockholm Business School): Why are European Equities so Illiquid?
Kiyoshi Kanazawa (Kyoto University): A census study of order splitting and the square-root price impact law on the Tokyo Stock Exchange
The long memory of market-order flow is an established stylised fact in financial market microstructure. This empirical fact is theoretically hypothesised by the Lillo-Mike-Farmer (LFM) model, which argues that metaorder splitting behaviour of individual traders is the key to understanding the long memory. Also, it is empirically claimed that the price impact of these metaorders is characterised by the square-root scaling law. Here, we have two research questions: (1) If the LMF model is correct, the autocorrelation function (ACF) of market-order signs can be quantitatively predicted. Is such an LMF prediction valid even at the quantitative level? (2) There has been a debate about the scaling exponent δ in the square-root law. Some researchers claim that δ is exactly equal to 1/2 for any assets, suggesting its universality. Others claim that δ depends on specific conditions of assets, suggesting its non-universality. Is δ universally equal to 1/2, or does it depend on asset-specific conditions? These questions have remained unresolved due to a lack of microscopic, census-level data on individual traders. In this talk, we show our data analytical results on the LMF hypothesis and the universality of the square-root law. We used a microscopic dataset on the Tokyo Stock Exchange (TSE). This dataset includes all the virtual server IDs - a unit of trading accounts on the TSE - and enables us to study the trading behaviour of all traders effectively. Our conclusions are summarised as follows: (1) The LMF prediction was valid even quantitatively for the market-order sign ACF. (2) The power-law exponent δ in the square-root law was 1/2 for all stocks on the TSE within statistical errors. This result suggests the strict universality of the square-root law.
Phil Mackintosh (Nasdaq): Fragmentation and Segmentation is all Economics
Since the late 90’s markets have automated and regulators have focused on competition.
We show that economics of trading reward more fragmented markets and increasingly segmented markets. Often regulators have unintentionally created free-rider problems.
Public markets are good for a number of things – raising capital, reducing costs of trading and making markets more efficient. Extrapolating recent evolution – might be making public markets less effective and less attractive.
Stefan Schlamp (Deutsche Börse): HFT State of Play
Old-school HFT just meant servers collocated with the exchange. Then came dedicated fiber optic connections between trading venues (e.g. the “Spread Line” between Chicago and New York) which were soon replaced by series of microwave towers. The latest development is the use of short-wave radio transmitters to send signals across the Atlantic and Pacific.
Within the collocated servers, SolarFlare network cards with kernel bypass became a condition sine qua non which were soon made obsolete by FPGAs. These allowed reaction times to events of less than 100 nanoseconds employing rule-based methodologies. The current state of the art uses ASICs and exploits the subtle details of the data transmission protocols to yield wire-to-wire reaction times in the low single-digit nanosecond range!
The talk will go over some of these developments, their impact on non-HFT participants, their prevalence and fingerprints in the market data, and how exchanges can/do/should respond.
Justin Sirignano (University of Oxford): Real-time Recurrent Learning for Recurrent Neural Networks: Convergence Analysis and Applications to High-Frequency Data
Recurrent neural networks (RNNs) are commonly trained with the truncated backpropagation-through-time (TBPTT) algorithm. For the purposes of computational tractability, the TBPTT algorithm truncates the chain rule and calculates the gradient on a finite block of the overall data sequence. Such approximation could lead to significant inaccuracies, as the block length for the truncated backpropagation is typically limited to be much smaller than the overall sequence length.
In contrast, real-time recurrent learning (RTRL) is an online optimization algorithm which asymptotically follows the true gradient of the loss on the data sequence as the number of sequence time steps tends to infinity. RTRL forward propagates the derivatives of the RNN hidden/memory units with respect to the parameters and, using the forward derivatives, performs online updates of the parameters at each time step in the data sequence. RTRL's online forward propagation allows for exact optimization over extremely long data sequences, which makes it potentially appealing for long data sequences (such as order book data in finance), although it can be computationally costly for models with large numbers of parameters.
We prove convergence of the RTRL algorithm for a class of RNNs using a fixed point analysis combined with a Poisson equation to appropriately bound the fluctuations of the RNN hidden layer during training. We will conclude by presenting several applications of RTRL for modeling order book data, stochastic optimal control, and reinforcement learning in finance, comparing its performance against TBPTT.
Almut Veraart (Imperial College London): TBD
Dario Villamaina (Capital Fund Management): Self-Inflated Funds
When funds with illiquid portfolios grow rapidly without rebalancing into more liquid assets, they generate self-inflated returns via their own price impact. Investors chase these returns, triggering a positive feedback loop that inflates both fund size and asset prices.
We introduce a simple measure - fund illiquidity - that captures a fund’s potential for return inflation. Using daily ETF data, we estimate price impact, decompose returns, and show that investors chase both fundamental and self-inflated returns. We find that inflated funds underperform in the long run and that stock-level ownership by inflated funds predicts negative long-term returns.
Ji Hee Yoon (University College London): Market Choice in Asset Trading: The Role of Collateral
Many financial assets are simultaneously traded on centralized exchanges and on bilateralover-the-counter (OTC) markets. We investigate investors’ decisions to enter one of these two markets, with a specific focus on collateral requirements, which are imposed by many exchanges in practice.
We find that in equilibrium OTC markets and exchanges coexist when the search technology for an OTC counterparty is sufficiently liquid and transparent. The trade-off between the collateral constraint and search frictions tends to encourage investors with large trading needs to enter the OTC market and those with small trading needs to trade on the exchange. When the search procedure is not fully transparent, however, the presence of investors with large trading needs on the OTC market may also attract investors with small trading needs (speculators) to that market who are aiming to profit from the possibility of matches with investors with large trading needs. We then find that in such an equilibrium there is too much entry in the OTC market, and we show that welfare could be increased by taxing entry to that market.