Market Map

  • 114 ETFs selected from source across the entire canvas shown on finviz.
  • Uses intraday real time data, updating every 5 seconds from 9:30 AM to 4:00 PM EST.
  • Natural log first differences of the raw price series are used to estimate the return.
  • Correlation Graph/Network – This high dimensional matrix is presented in a 2D plane for clarity. Correlations are measured from 9:30 AM EST, the data frame is updated every 5 seconds, anchored at 9:30 AM EST.
  • Time Series Data – Displays the 5 second return of each ETF on the figure’s right side, natural log of the price differences, ln(T(0)) – ln(T(-1)) are plotted via spark lines.
  • Projection Technique – Utilizes the Sammon’s projection for mapping ETF’s onto a 2D plane, emphasizing the importance of spatial location. This method results in an ETF embedding space. ETFs positioned closer together exhibit more similar behaviors than those further apart, with respect to the correlation of itself to all other ETFs.
  • Illustrated by a row of circles at the bottom, representing ongoing Factor analysis with varimax rotation.
  • The factors are sorted by the number of variables it explains (eigenvalue), offering insight into market dominance throughout the day. Think of these as strings on a musical instrument, number of dominant strings active for the day and how loud they are vibrating.
  • Volatility and Market Factors – The shape of the network changes with market volatility, becoming denser on high-volatility days. Some days 1 factor explains over 50% of the total market variance, usually under extreme stress, i.e. a selloff, or a melt-up.
  • The map, due to being an EMBEDDING, will show clear groups, if they exist. Clusters will appear on regions of the map based on type of asset, like equity, volatility, negative correlation products, debt, commodity, currency for example. Anomalies or outliers can be observed by viewing an ETF’s movement on the graph over the day as well its location in the embedding.
  • Color-coded to indicate positive (green) or negative (red) correlations.
  • Line thickness signifies the strength or degree of correlation between ETFs.
  • Hovering over a node reveals other ETFs that are correlated as well as the sorted dominant factor observed for the day.
  • Explore by clicking the left gutter for filters, right gutter for the entities or nodes in the 2D graph drill-downs, the canvas itself has tabs running vertically for different data science methods being run in the background.
  • Left gutter enables filtering on different correlation strengths, as well as taking the absolute value if required for the filter to be applied.

Tail Hedge

Figure 2: Sample output from the Tail Hedge Simulator is focused on calculating put prices when a large swift decline occurs in an ETF’s simulated future. Inspired by tail hedge strategies, Black Swan events, Spitznagel and Mandelbrot authors.

Example Use Case –
“Right now, what market PUT option would provide me the most left tailed convexity assuming 30 days into the future a 20% percent decline will occur in QQQ”. In other words, examine all the actual QQQ expiration date CROSS strike price PUT option combinations post expiry T + 30 days. What are each combination’s respective return estimates based on a forward walk simulation. Return is defined as the current market ASK PUT price vs. the expected MAX of that PUT option price in the 20% decline period. This aligns with finding convex vehicles or explosive payoff combinations, if they exist.

“I now want to simulate how many PUT contracts I need to purchase to achieve a 100% hedge when QQQ has an expected 20% decline”. My total QQQ long position $ net change with the hedge vs. the # of contracts, a plot and table would be great to help me make a hedge coverage vs. cost decision.

This summarizes the use cases for the Tail Hedge Simulator.

Macro-Economic Forecast

Figure 3: Sample output of the Ex-Post forecast from the Macro-Economic Forecast tool.

The Macro-Economic Forecast tool analyzes various macro-economic indicators using a Vector Autoregression (VAR) model. By default, the model is trained on historical data from January 1998 to December 2023, incorporating 14 lags. Both the time periods and number of lags can be customized.

This analysis uses historical data procured from various sources to predict Coincident Indicator (COIN), 10-Year Treasury Note Yield, S&P 500 ETF, Unemployment Rate, Yield Curve, Consumer Price Index: All Items Less Food & Energy(CPI), Copper ETF, Spot Oil Price:West Texas Intermediate, Risk Premium and US Dollar Index macro-economic indicators.

  • Ex-Post Forecasting: Ex-post forecasting is a hindsight forecast for a period that has already happened. Ex-Post forecast enables a thorough assessment of the model’s accuracy over the selected testing period.
  • Ex-Ante Forecasting: Ex-Ante forecasting is a prediction about the future made with information available at present. This is a forecast done entirely on estimates, without the benefit of knowing the actual values.
  • Time Period Selection: Time periods of past data for training and testing the forecasting model can be customized. This will be used by the model to make predictions based on the trends observed in that duration.
  • Number of Lags: The number of previous observations used as predictors in a time series analysis or forecasting model. The number of lags can be adjusted to fine-tune the model’s sensitivity to macro-economic dynamics.