Currency Trading

Does the increase in number of Bitcoin wallets at a rate that far exceeds growth in number of Bitcoins explain the dramatic rise in Bitcoin price? In the December revision of his paper entitled “Metcalfe’s Law as a Model for Bitcoin’s Value”, Timothy Peterson models Bitcoin price according to Metcalfe’ Law, which posits that the value of a network (Bitcoin) is a function of the number of possible pair connections (among Bitcoin wallets, assuming all are equal) and is therefore proportional to the square of the number of participants. Said differently, he models Bitcoin value based on supply (number of Bitcoins) and demand (number of Bitcoin wallets). Per Metcalfe’s Law, Bitcoin return is proportional to twice the growth rate of Bitcoin wallets. He tests the model via a least squares regression of actual Bitcoin price on modeled price with adjustment for inflation due to new Bitcoin creation. He applies the model to investigate claims of Bitcoin price manipulation during 2013-2014. Using number of Bitcoins and number of Bitcoin wallets at 60-day intervals during December 31, 2011 through September 30, 2017, he finds that:

Keep Reading

Are there rational ways to decide whether cryptocurrencies such as Bitcoin are in bubbles? In their December 2017 paper entitled “Datestamping the Bitcoin and Ethereum Bubbles”, Shaen Corbet, Brian Lucey and Larisa Yarovaya test for bubbles in Bitcoin and Ethereum price series. For valuation, they consider three potential cyrptocurrency price drivers:

1. Blockchain length, reflecting difficulty of finding a new block and receiving payment relative to past difficulty. As more miners engage, the rate of block creation increases, raising the level of difficulty.
2. Hash rate, indicating speed of blockchain code execution during mining. A higher hash rate increases probability of finding the next block and receiving payment. 
3. Liquidity, measuring the relationship between cryptocurrency daily returns and volatilities. 

They then apply ratios constructed from these variables to detect times when price series are substantially disconnected from fundamental drivers. Using Bitcoin data since July 18, 2010 and Ethereum data since July 30, 2015, both through November 9, 2017, they find that: Keep Reading

Are cryptocurrencies potentially useful portfolio diversifiers? In their November 2017 paper entitled “Exploring the Dynamic Relationships between Cryptocurrencies and Other Financial Assets”, Shaen Corbet, Andrew Meegan, Charles Larkin, Brian Lucey and Larisa Yarovaya apply a battery of tests to analyze relationships: (1) among three cryptocurrencies; and, (2) between the cryptocurrencies and conventional asset classes. They consider cryptocurrencies with market values over $1B at the end July 2017: Bitcoin, Ripple and Litecoin. They consider equities (S&P 500 Index), bonds (Markit ITTR110), commodities (S&P GSCI Total Returns Index), currencies (U.S. Dollar Broad Index), gold (COMEX close) and S&P 500 implied volatility (VIX) as conventional asset classes. Using daily data for Bitcoin, Ripple and Litecoin and for conventional asset classes as specified during April 29, 2013 through April 30, 2017, they find that: Keep Reading

Is value investing particularly profitable when the price spread between cheap and expensive assets (the value spread) is extremely large (deep value)? In their November 2017 paper entitled “Deep Value”, Clifford Asness, John Liew, Lasse Pedersen and Ashwin Thapar examine how the performance of value investing changes when the value spread is in its largest fifth (quintile). They consider value spreads for seven asset classes: individual stocks within each of four global regions (U.S., UK, continental Europe and Japan); equity index futures globally; currencies globally; and, bond futures globally. Their measures for value are:

• Individual stocks – book value-to-market capitalization ratio (B/P).
• Equity index futures – index-level B/P, aggregated using index weights.
• Currencies – real exchange rate based on purchasing power parity.
• Bonds – real bond yield (nominal bond yield minus forecasted inflation).

For each of the seven broad asset classes, they each month rank assets by value. They then for each class form a hedge portfolio that is long (short) the third of assets that are cheapest (most expensive). For stocks and equity indexes, they weight portfolio assets by market capitalization. For currencies and bond futures, they weight equally. To create more deep value episodes, they construct 515 sub-classes from the seven broad asset classes. For asset sub-classes, they use hedge portfolios when there are many assets (272 strategies) and pairs trading when there are few (243 strategies). They conduct both in-sample and out-of-sample deep value tests, the latter buying value when the value spread is within its top inception-to-date quintile and selling value when the value spread reverts to its inception-to-date median. Using data as specified and as available (starting as early as January 1926 for U.S. stocks and as late as January 1988 for continental Europe stocks) through September 2015, they find that:

Keep Reading

Do value strategy returns vary exploitably over time and across asset classes? In their October 2017 paper entitled “Value Timing: Risk and Return Across Asset Classes”, Fahiz Baba Yara, Martijn Boons and Andrea Tamoni examine the power of value spreads to predict returns for individual U.S. equities, global stock indexes, global government bonds, commodities and currencies. They measure value spreads as follows:

• For individual stocks, they each month sort stocks into tenths (deciles) on book-to-market ratio and form a portfolio that is long (short) the value-weighted decile with the highest (lowest) ratios.
• For global developed market equity indexes, they each month form a portfolio that is long (short) the equally weighted indexes with book-to-price ratio above (below) the median.
• For each other asset class, they each month form a portfolio that is long (short) the equally weighted assets with 5-year past returns below (above) the median.

To quantify benefits of timing value spreads, they test monthly time series (in only when undervalued) and rotation (weighted by valuation) strategies across asset classes. To measure sources of value spread variation, they decompose value spreads into asset class-specific and common components. Using monthly data for liquid U.S. stocks during January 1972 through December 2014, spot prices for 28 commodities during January 1972 through December 2014, spot and forward exchange rates for 10 currencies during February 1976 through December 2014, modeled and 1-month futures prices for ten 10-year government bonds during January 1991 through May 2009, and levels and book-to-price ratios for 13 developed equity market indexes during January 1994 through December 2014, they find that:

Keep Reading

Does low volume in currency exchange markets expose exploitable inefficiencies? In their August 2017 paper entitled “The Value of Volume in Foreign Exchange”, Antonio Gargano, Steven Riddiough and Lucio Sarno investigate whether currency trading volumes (including spot, swap and forward) exploitably predict currency returns. They first measure interactions of trading volumes and returns statistically. They then assess gross economic import via portfolios formed from daily double-sorts first on prior-day returns and then on prior-day trading volumes, focusing on a portfolio that is each day long (short) currency pairs with low (high) prior returns and low trading volumes. Finally, they incorporate bid-ask spreads to determine whether net portfolio performance is attractive. Using hourly spot, swap and forward trading volumes and their daily returns across 31 currency pairs during November 2011 through December 2016, they find that:

Keep Reading

Does a carry trade derived from roll yields of futures/forward contracts work within asset classes (undiversified) and across asset classes (iversified)? In his May 2017 paper entitled “Optimising Cross-Asset Carry”, Nick Baltas explores the profitability of cross-sectional (relative) and time-series (absolute) carry strategies within and across futures/forward markets for currencies, stock indexes, commodities and government bonds. He posits that contracts in backwardation (contango) present a positive (negative) roll yield and should generally be overweighted (underweighted) in a carry portfolio. He considers three types of carry portfolios, each reformed monthly:

1. Cross-sectional (XS) or Relative – Rank all assets within a class by strength of carry, demean the rankings such that half are positive and half are negative and then assign weights proportional to demeaned ranks to create a balanced long-short portfolio. Combine asset classes by applying inverse volatility weights (based on 100-day rolling windows of returns) to each class portfolio.
2. Times-series (TS) or Absolute – Go long (short) each asset within a class that is in backwardation (contango), such that the class may be net long or short. Combine asset classes in the same way as XS.
3. Optimized (OPT) – Apply both relative strength and sign of carry to determine gross magnitude and direction (long or short) of positions for all assets, and further apply asset volatilities and correlations (based on 100-day rolling windows of returns) to optimize return/risk allocations.

Using daily data for 52 futures series (20 commodities, eight 10-year government bonds, nine currency exchange rates versus the U.S. dollar and 15 country stock indexes) during January 1990 through January 2016, he finds that: Keep Reading

How profitable is automated multi-horizon extraction of liquidity premiums in currency exchange markets? In their April 2017 paper entitled “The Alpha Engine: Designing an Automated Trading Algorithm”, Anton Golub, James Glattfelder and Richard Olsen introduce an adaptive counter-trend algorithmic trading system that seeks liquidity premiums from price series via automated trades at adaptive market events. The system consists of the following building blocks:

• Module that employs an event-based (intrinsic) time scale to determine price series directional changes and overshoots. 
• Module that analyzes relationships between price series directional changes and overshoots over multiple (baseline four) horizons.
• Module that sorts directional changes (upward or downward) to enable asymmetric overshoot thresholds.
• Module that trades at empirically adaptive events.
• Module that sizes trades by identifying degrees to which associated market conditions are abnormal.
• Module that suppresses accumulation of large inventories during long market trends.

For opportunity generation and execution, they require: intraday trading capability; full automation; and, limit orders (to the extent possible). They illustrate the system on currency trading. Using intraday data for 23 currency exchange rates during 2006 through 2013, they find that: Keep Reading

How well do time series (intrinsic) and cross-sectional (relative) momentum work for different types of currency exchange rates? In their April 2017 paper entitled “Momentum in Traditional and Cryptocurrencies Made Simple”, Janick Rohrbach, Silvan Suremann and Joerg Osterrieder compare the effectiveness of time series and cross-sectional momentum as applied to three groups of currency exchange rates: G10 currencies; non-G10 conventional currencies; and, cryptocurrencies. To measure momentum they employ three pairs (one fast and one slow) of exponential moving averages (EMA) spanning short, intermediate and long horizons. When the fast EMA of a pair is above (below) the slow EMA, the trend is positive (negative). They extract a momentum signal for each exchange rate from these three EMA pairs by:

1. For each EMA pair, taking the difference between the fast and slow EMA.
2. For each EMA pair, dividing the output of step 1 by the standard deviation of the exchange rate over the last three months to scale currency fluctuations to the same magnitude.
3. For each EMA pair, dividing the output of step 2 by its own standard deviation over the last year to suppress series volatility.
4. For each EMA pair, mapping all outputs of step 3 to signals between -1 and 1.
5. Averaging the signals across the three EMA pairs to produce an overall momentum signal.

The time series portfolio holds all currencies weighted each day according to their respective prior-day overall momentum signals.  The cross-sectional portfolio is each day long (short) the three currencies with the highest (lowest) overall momentum signals. Key performance metrics are annualized average gross return, annualized standard deviation of returns, annualized gross Sharpe ratio (assuming risk-free rate 0%) and maximum drawdown. Using daily foreign currency exchange rates for 23 conventional currencies and seven cryptocurrencies versus the U.S. dollar as available through late March 2017, they find that: Keep Reading

Are anomaly premiums (expected winners minus losers among assets within a class, based on some asset characteristic) more or less predictable than broad market returns? In their April 2017 paper entitled “Predicting Relative Returns”, Valentin Haddad, Serhiy Kozak and Shrihari Santosh apply principal component analysis to assess the predictability of premiums for published asset pricing anomalies spanning stocks, U.S. Treasuries and currencies. For tractability, they simplify asset classes by forming portfolios of assets within them, as follows:

• For stocks, they consider the long and short legs of portfolios reformed monthly into tenths (deciles) based on each of the characteristics associated with 26 published stock return anomalies (monthly data for 1973 through 2015).
• They sort zero-coupon U.S. Treasuries by maturity from one to 15 years to assess term premiums (yield data for 1985 through 2014).
• They sort individual exchange rates into five portfolios reformed daily based on interest rate differentials with the U.S. to assess the carry trade premium (daily data as available for December 1975 through December 2016).

Using the specified data, they find that: Keep Reading