Side StreetResearch

Research / The scientific program

The economics
behind the pattern.

Our organizing question: how do capital, competition, information and implementation constraints determine the efficiency of financial markets?

THREE PROGRAMS
SIX CONNECTED LENSES
01 / A connected agendaLong-term research direction
01

Efficiency & capacity

How does feasible scale change a discrepancy?

Capital occupancy
02

Cross-market price discovery

Which venue informs the others?

Information transmission
03

Liquidity & microstructure

What does a received quote change reveal?

Adverse selection
04

Derivatives & financing

What keeps capital from converging prices?

Fragmented collateral
05

Nonlinear market dynamics

When does the same mechanism change regime?

Liquidity feedback
06

Computational discovery

Can AI improve verified research throughput?

Controlled hypothesis search

Bitcoin is our empirical starting point. The wider program is intended to span instruments, venues, liquidity regimes and horizons. These programs are research objectives; they are not established discoveries.

02 / Priority programsOpen a program to examine its design

Priority research programs

01
Research direction

The Efficiency–Capacity Frontier

How much capital can an opportunity accommodate?

The mechanism

Information, financing and execution have costs. An apparent discrepancy can be attractive at one size and disappear as depth is consumed, capital is occupied or risk increases.

Founding theory
Costly information and limits to arbitrage motivate the question. They do not establish that smaller markets offer profitable opportunities.
Grossman & Stiglitz (1980) ↗Shleifer & Vishny (1997) ↗Makarov & Schoar (2020) ↗
The empirical question
Do independently measured access, financing and liquidity constraints explain residual price discrepancies—and their feasible size—beyond information and risk?
Proposed evidence
A predefined cross-venue instrument panel; comparable contracts and clocks; observed depth, fees and financing terms; a frozen size grid. Compare prefunded inventory with transfer-dependent routes and retain failed legs, missing observations and capital occupancy.
The decision rule
Advance only if conservative net-feasibility estimates survive matched risk and liquidity controls and later chronology. Nonpositive value at the smallest feasible size rejects the proposed trading route.
What it could contribute
A reproducible taxonomy connecting opportunity scale with the economic constraints that shape convergence.
02
Research direction

The Anatomy of Price Discovery

Where does information enter, and how does it travel?

The mechanism

Related markets do not receive identical information or supply identical liquidity. Order flow, adverse selection and fragmented access can affect the path of price adjustment.

Founding theory
Kyle and Glosten–Milgrom connect trading to information and liquidity. Hasbrouck provides a framework for contributions to price discovery; historical cryptocurrency studies motivate renewed measurement.
Kyle (1985) ↗Glosten & Milgrom (1985) ↗Hasbrouck (1995) ↗Makarov & Schoar (2020) ↗
The empirical question
When does a venue contribute new information, and when does it merely display a stale or mechanically delayed response? How does that distinction change with liquidity and market activity?
Proposed evidence
Synchronized, instrument-comparable observations; explicit receive times and quote validity; information-share and lead-lag comparisons with clock, activity and common-shock controls. Native quote-response measurements require separate qualification before prediction.
The decision rule
Reject an informational interpretation if the effect disappears after quote-age, timing and simpler price/activity controls. Estimated leadership alone does not demonstrate executable arbitrage or causality.
What it could contribute
Measurement frameworks that distinguish genuine information transmission from the artifacts of fragmented observation.
03
Research direction

The Economics of Execution

Where does a forecast lose its economic value?

The mechanism

A forecast becomes a position through orders, fills and financing. Delay, spreads, impact, inventory and uncertain execution can consume its apparent value.

Founding theory
Almgren–Chriss makes impact and timing risk explicit. Adverse-selection models explain why marked price changes and realizable trading outcomes can differ.
Almgren & Chriss (2001) ↗Glosten & Milgrom (1985) ↗Bailey et al. (2017) ↗Harvey, Liu & Zhu (2016) ↗
The empirical question
Which implementation constraint explains the greatest loss of value between a statistically informative forecast and a funded, risk-constrained decision?
Proposed evidence
A preselected rule with fixed entry, exit and abstention conventions; size, delay, fee and depth sensitivities; explicit partial-fill, adverse-selection and financing assumptions. Separate exogenous-book replay from observed fills and endogenous impact.
The decision rule
Require economic value beyond simpler controls, credible cost assumptions and independent replication before prospective evaluation. A favorable gross-price calculation is insufficient.
What it could contribute
A common accounting framework linking predictions to execution, capital use and the decision to continue, revise or reject a strategy.
03 / Focused investigationsAdditional lenses · proposed studies

Focused investigations

04
Research proposal

Financing, collateral and the basis

The mechanism

Collateral cannot always move freely between venues. Borrow, margin, settlement and credit exposure can delay convergence or explain a premium as compensation for risk.

Founding theory
Limits to arbitrage and historical market segmentation motivate a distinction between quoted basis and realized, funded carry.
Shleifer & Vishny (1997) ↗Makarov & Schoar (2020) ↗
The empirical question
Do measured financing and access constraints explain residual dislocation persistence beyond liquidity, volatility and information?
Proposed evidence
Comparable spot–derivative prices, realized funding, independently observed borrow and collateral terms, and versioned rules. Use matched windows around measured friction changes; test pretrends before any causal interpretation.
What it could contribute
A constraint taxonomy and capital-occupancy accounting. A nominal basis or a delta hedge does not make a position riskless.
05
Research proposal

Liquidity response across regimes

The mechanism

Liquidity can retreat as volatility, forced flow or inventory pressure rises. The same observation may have different implications across participation and funding conditions.

Founding theory
Adverse selection and history-dependent order-book impact motivate questions about liquidity adaptation. A nonlinear model alone does not identify feedback.
Glosten & Milgrom (1985) ↗Eisler, Bouchaud & Kockelkoren (2012) ↗
The empirical question
Does a prospectively defined regime change the incremental information or net feasibility of a fixed mechanism?
Proposed evidence
Frozen regime definitions, event streams and common-news controls; matched activity and price baselines; later-period comparisons. Retain stress-period missingness and scarce support rather than selecting favorable episodes.
What it could contribute
A framework for regime-dependent price formation that distinguishes descriptive response, omitted common shocks and genuine incremental evidence.
06
Research proposal

Discovery with a controlled search budget

The mechanism

AI can reduce synthesis and implementation effort while introducing correlated hypotheses, source errors and defective code. Useful throughput depends on accepted work and the cost of correction.

Founding theory
Multiple-testing and backtest-overfitting research make the search process part of the evidence. Faster idea generation can increase selection risk.
Bailey et al. (2017) ↗Harvey, Liu & Zhu (2016) ↗
The empirical question
Can bounded AI assistance improve reproducible research quality and throughput without expanding false discoveries?
Proposed evidence
A prospective logged task comparison with matched access and resources, blinded review, deterministic checks and retained rejections. Measure accepted artifact time, defects, source accuracy and correction burden.
What it could contribute
Supervised hypothesis workflows judged by verified deliverables, not autonomy claims. This comparative evaluation is planned; AI assistance already has a bounded supporting role.
04 / Research notesDocumented Side Street work

The record is a starting point.
Not the ceiling.

The following notes summarize completed exploratory Bitcoin studies and bounded engineering work. They describe research decisions, rather than a trading track record.

R.01Completed · exploratory

Predicting movement is not predicting direction.

Chronological Bitcoin comparisons separated event occurrence from conditional direction using proper probability scores. Several occurrence improvements coexisted with weak or unstable signed forecasts. These were previously explored development periods; the result is a diagnostic finding, not independent confirmation or trading performance.

Research implicationMeasure the information needed for the decision, rather than treating a better aggregate score as an economic result.

R.02Completed · negative finding

A complex hypothesis still has to beat a simple explanation.

Control-first comparisons challenged short-horizon ranking, target-allocation and ordered post-event representations against simpler alternatives. The tested recipes failed their declared advancement criteria. Rare continuation outcomes and temporal concentration limited stronger claims; the findings do not reject every temporal or microstructure mechanism.

Research implicationRetire the tested recipe when controls fail. More model complexity is not an automatic remedy.

R.03Qualified measurement · prediction pending

Before prediction, establish what is observable.

A received-message observable of net displayed quote response was qualified on a bounded sample, alongside a feature-only event calendar. Arithmetic, witnesses and population identities were checked. This does not identify hidden liquidity or participant intent. Native streaming materialization and the incremental predictive comparison remain planned.

Research implicationMeasurement qualification is a distinct deliverable. It cannot stand in for predictive or economic validation.

Earlier work also examined order-flow sign, queue imbalance, depth dynamics and derivatives-linked state. Results were negative, inconclusive or descriptive within their specified comparisons; no validated net-profit or capacity claim followed.

05 / Intellectual foundationsPublished theory and historical evidence

A foundation, not a borrowed conclusion.

These sources inform mechanisms and research design. Stylized models and historical findings do not prove our proposed empirical thesis.

  1. Grossman & Stiglitz1980
    On the Impossibility of Informationally Efficient Markets

    The incentives required to produce costly information.

  2. Shleifer & Vishny1997
    The Limits of Arbitrage

    Risk and delegated-capital constraints. Linked working-paper version: 1995.

  3. Kyle1985
    Continuous Auctions and Insider Trading

    Informed order flow, market depth and price adjustment.

  4. Glosten & Milgrom1985
    Bid, Ask, and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders

    Adverse selection and the difference between observed and realizable returns.

  5. Hasbrouck1995
    One Security, Many Markets: Determining the Contributions to Price Discovery

    An econometric framework for linked-market information shares.

  6. Almgren & Chriss2001
    Optimal Execution of Portfolio Transactions

    The trade-off between execution speed, impact and risk.

  7. Makarov & Schoar2020
    Trading and Arbitrage in Cryptocurrency Markets

    Historical evidence on segmentation. Linked institutional working-paper version; not evidence of present opportunities.

  8. Bailey et al.2017
    The Probability of Backtest Overfitting

    Selection risk across strategy trials; not a substitute for chronological confirmation.

  9. Harvey, Liu & Zhu2016
    …and the Cross-Section of Expected Returns

    Multiple testing and the evidence required for return-pattern claims.

  10. Eisler, Bouchaud & Kockelkoren2012
    The Price Impact of Order Book Events: Market Orders, Limit Orders and Cancellations

    History-dependent order-book impact in equity data. Linked author preprint; no crypto parameters are inferred.

How these questions become tests