Over the course of more than 50 years of combined commodity trading experience, Martin Fund Management’s (MFM) CIO David Martin and Director of Business Development, Paul Wankmueller, have seen multiple cycles including bull, bear and rangebound markets for individual markets and commodity super-cycle bull markets across the asset class. Their institutional long/short commodity trading strategy, launched in May 2020, has generated a Sharpe ratio above one and a Sortino ratio above two, annualized at around 30% net of fees to June 2023, and earned several performance awards. The directional strategy is substantially systematic, but is not correlated to trend following CTAs nor other systematic macro strategies.
Since setting up MFM to run third party capital in 2013, Martin has received overtures from the giant multi-strategy platforms, which would appreciate the diversification benefits of his strategy, but he prefers to remain independent. He spent two million dollars setting up the firm, which is CFTC and NFA registered. Current prime brokers are Goldman Sachs and Société Generale. SS&C Eze software is used for institutional quality portfolio management, order management, execution etc. There are also multiple Bloomberg terminals.
We trade with wide brush strokes, using a house painting brush and not a fine scalpel.
David Martin, CIO, Martin Fund Management
Martin was previously a discretionary trader of directional and relative value strategies, using proprietary capital, at the New York Board of Trade and the New York Mercantile Exchange. Over the years his approach has evolved and adapted to the market, broker and venue environments. Higher exchange fees, regulatory fees and commissions have wiped out the profitability of scalping a few ticks, and raised the return hurdle for some relative value trades such as calendar spreads, which have also become more crowded. This has increased the incentives to try and capture larger directional moves, and carefully crafted option structures are well suited to making such wagers with limited risk. Martin has also determined that a rules-based approach is optimal, and systematized the insights from his discretionary trading into a framework and process that takes some emotion and discretion out of the strategy: “We have tight predetermined risk parameters, which systematize trade entry and exit. Trailing stops can also kick in,” says Martin. The process is primarily systematic, but can exercise discretion in areas such as reducing risk and sizing based on conviction.
Signal generation models blend technical and fundamental inputs using a proprietary scoring and weighting system. Technicals include well known indicators such as support, resistance, moving averages, standard deviations and volatility levels. Fundamental supply and demand data includes harvest and crop estimates, demand, seasonal factors and inventory-related data, such as certified exchange stocks, storage and warehouse receipts. Cost of carry is also monitored, and costs of production are modelled in USD and local currency. The systems also aim to distinguish human from algorithmic trading, monitoring key market players including macro funds, CTAs and index funds. MFM has also developed proprietary methods of ranking and scoring sentiment and positioning indicators based on open interest, commitment of traders and other metrics. Notwithstanding some proprietary data, many of the inputs are fairly well known and widely used, but the secret sauce is the recipe of how they are weighted and combined.
The data is synthesized into a bullish, bearish or rangebound forecast. The individual trades could sometimes be trend following or countertrend, but Martin does not monitor this split because he thinks that doing so could introduce a cognitive bias, which is one form of behavioural bias heeded by the models. MFM analyses the fundamentals but is also aware that market behaviour can deviate from economic logic. For instance, in January 2023, Martin judged that, “Coffee around 1.40 to 1.45 per pound is not far from aggregate costs of production so has limited downside because lower prices would reduce supply; equally we recognize that momentum traders could continue selling it down”.
The formulas for the models were originally written in Excel, but have been converted into Bloomberg Python coding language using a machine learning application, API XG Boost. The package is open source, but the way in which MFM uses it is proprietary. Martin’s interest in logic and programming dates back to his Wharton degree in the late 1980s, which focused on Decision Sciences, decision trees, logic, game theory and database management. Wankmueller, who is an experienced market technician, a CMT (chartered market technician), and sits on the editorial board of the Journal of Technical Analysis, has also contributed to the system development. The duo has additionally developed a strong network of relationships with brokers and commodity producers, ranging from family cooperatives in Brazil to traders in the London metal markets.
MFM constructs premium neutral asymmetric option structures with limited downside risk, but unlimited upside. This avoids the negative convexity of a linear delta one approach, or some other sorts of option structures such as risk reversals or certain types of ratio spreads.
Typically, for a bullish trade, MFM will buy a 20-delta call option and finance it by selling one or two 10 delta put spreads. For instance, when Arabica coffee was trading around 130 in early 2021, MFM bought a 170 strike, 20 delta call, and sold put spreads at 110/100, paying for the call. This trade hit a home run as coffee rallied close to 2.50. For a bearish trade, Martin would buy a put and sell a call spread, with similar deltas.
The best month for the strategy so far, December 2020, saw a net return of 22.25%, well above target.
When working at Cooper Neff & Associates, Martin learnt about the “Greeks” or option sensitivities, which are all monitored, and can feed into strike selection and trade structuring, but the Greeks are not so relevant to the profit and loss on the option structures, which are mainly geared towards expressing directional views and capturing the deltas. Admittedly, Martin has sometimes seen skews reach extreme levels, which could influence relative pricing of the option structures’ legs, but the directional move is expected to be the main return driver.
If rangebound price action is expected, an iron condor could be constructed. This sells an out of the money call spread and put spread, with the aim of collecting a fixed payout, also with limited risk. “This is a delta neutral and market neutral structure designed to collect premium,” says Martin.
Six-month options are generally selected as a sweet spot tenor, balancing the need to earn premium income on shorts to cover long option costs with the need to maintain liquidity. “A 12-month option would earn more income but would also be less liquid,” points out Martin. The average holding period is 3 months because some structures are exited before maturity. Equally exposures can continue beyond 6 months: if the models signal a continuing trend a new option structure is opened after the old one expires.
The trade allocation is completely opportunistic, so the portfolio could, in theory, be wholly exposed to bullish, bearish or rangebound trades. There is no minimum weight for any of the three. “That said, we are always long the wings and positioned to profit from the far ends of the price spectrum,” says Martin.
Since 1990, he estimates his hit rate is about 70% winners – and the win loss ratio is also favourable: “We trade with wide brush strokes, using a house painting brush and not a fine scalpel,” says Martin.
Electric vehicle infrastructure such as charging stations is not yet rolled out. This is a bullish factor for copper, natural gas and platinum... but this needs to be balanced against recessionary fears.
David Martin, CIO, Martin Fund Management
The best month for the strategy so far, December 2020, saw a net return of 22.25%, well above target. “We do not expect to make 20% in a month, but multiple calls moved well into the money,” recalls Martin. There was also some skillful tactical trading: the strategy captured profits from gold’s rally in December 2020, and then flipped to a short stance in January 2021.
The worst month was October 2022, down 16%, which saw a globally synchronized selloff in nearly all assets, including commodity markets such as cocoa, coffee and soybeans that are not normally correlated and have no intrinsic reasons to be correlated. The impact was higher because exposure was at the maximum of 6 positions. “It is generally rare to see correlation spikes between our diverse set of commodities,” says Martin.
This drawdown prompted MFM to introduce a correlation overlay filter, which monitors correlation spikes and can reduce risk, to curtail the quantum of losses in such scenarios. “The correlation filter would have reduced losses from 16% to 6% in October 2022, but would not have been activated on any other occasion since the program launched in May 2020,” confirms Martin.
Stop losses and trailing stops can also limit losses. A stop loss is defined as a maximum of 2% of fund or account NAV per trade structure from inception level, for a maximum conviction trade. After a run up in profit the distance from current market prices to the original stop can become larger than 2%. For instance, at inception, the stop losses across the maximum number of six top conviction positions could reach a peak of 12%, but if the fund is up 12% since then, the potential peak to trough drawdown, back down to stop loss levels could be 24% (if no trailing stops have yet been activated).
Once positions start moving into profit and hit a certain target, a trailing stop kicks in. “The trailing stops are reasonably wide to leave some headroom for price fluctuation in volatile markets. Trades are exited upon a retracement back to the trailing stop,” explains Martin.
(Trailing stops are used rather than simply rolling the option structures to a new set of strikes in order to avoid additional transaction costs.)
The worst-case loss per option structure at inception is 10%, though this would only occur with a huge gap in price action, gapping through the initial stop loss level. The worst loss per option structure is fixed at the intrinsic value of the short option spread plus the premium on the long option.
Various other risk constraints, including 20% maximum margin to equity, could also reduce position sizes at inception or later on.
Beyond MFM’s own risk reporting, bespoke risk reports are available from General Risk Advisors (GRA), founded by Ken Grant, who was previously Head of Global Risk Management for Cheyne Capital, Managing Director of Risk at SAC Capital Advisors and Director of Risk Management at Tudor Investment Corporation. Martin has been working with GRA for about 4 years, but not all investors require this level of detail.
Currently, MFM trades 7 markets: 3 softs (coffee, sugar and cocoa); 2 grains (corn and soybeans) and 2 metals (gold and copper). All of the options traded are priced in USD on the ICE and CME. Strategy capacity is estimated at USD 1 billion for these markets, based on liquidity rules of not owning over 10% of net speculative open interest in an option, which is determined from CFTC commitment of trader reports. “The net speculative open interest is defined as the net delta and historically the highest level we reached was only 8%. It is usually below 5%,” says Martin.
The manager has traded other markets over his career and continues to monitor a variety of commodity and other markets, which might be traded at some stage. If energy was added, capacity could be at least USD 3 billion and a new strategy expanding the universe to include Bitcoin and some equities might take it to USD 5 billion. Equally, some clients with managed accounts are pinpointing a narrower universe to reduce or avoid overlaps with their existing exposures.
Early investors opted for separately managed accounts but now most investors choose onshore and offshore funds for reasons including limited liability. The investor base has included some of the largest pension funds in the US, Germany and Switzerland, as well as corporate clients, family offices, and investment funds.
Though the strategy is systematic, Martin still has fundamental views and does see potential for another commodity Supercycle, partly caused by years of underinvestment in supply. “Commodity Supercycles generally last 8 to 15 years. The last one ran from 2000 to 2013. The new industrial revolution shifting from a petroleum-based world to one of green technology will require huge reindustrialization in the US and China. Electric vehicle infrastructure such as charging stations is not yet rolled out. This is a bullish factor for copper, natural gas and platinum, which could all be in huge demand. But this needs to be balanced against recessionary fears, due to higher interest rates. We do not see economic data supporting a recession yet,” says Martin.
The MFM strategy has potential to participate in the next commodity Supercycle but does not hold a religious level of conviction in the uber bullish thesis. Paying attention to technicals and sentiment means that MFM can also profit from down moves even when longer term fundamentals do seem constructive.