As part of the Trade Signal tournament, Metaculus is putting its money where its mouth is. Specifically, Metaculus forecasts developed within the tournament will be used to inform real trades with the $1,500 prize pool. This document aims to provide a summary of how this will be done.

We view this trading exercise as a pilot project with the primary goals of learning and having some fun. We have already learned quite a bit just from being forced to think through the operational details explained below. We’re eager to hear your suggestions for how our trading plan might be improved for future tournaments!

The Trading Theory

Our trading theory is conceptually quite simple. The basic idea is that market prices of stocks reflect investors’ current expectations of how the economy will develop over time. If economic indicators, like GDP, come in better than expected, then we generally expect the positive change in expectations to cause investors to broadly push US stock prices higher. For certain indicators, like industrial production, we expect changes in expectations to have the greatest impact on the stock prices of industrial companies.

If the Metaculus community were able to accurately forecast changes in investors’ economic expectations, then we would expect to be able to earn an extra return by buying stocks immediately before a better-than-expected GDP report and then selling immediately after the jump in stock prices, capturing the upward movement. Similarly, we could sell short stocks right before a worse-than-anticipated economic report, and benefit from an expected decline in stock prices.

Even under the best circumstances, we don’t expect our forecast will be more accurate than the market consensus every time. And, in cases when we do end up with a more accurate forecast than the market, we still don’t expect to make money on every trade, since the market’s reaction can be quite complex. However, if we can forecast economic indicators better than the market, on average, and by a big enough margin, then we should expect to be able to produce an extra investment return over time.

Translating the Theory Into Investments

The table below summarizes investments that we expect to be highly correlated with the various indicators in the Trade Signal tournament. For instance, we expect US building permits to be correlated with real estate stocks. If a particular month’s US building permit indicator is higher than the market expected, then we expect real estate stocks to increase.

$$\begin{array}{||c|c||} \hline \hline \textbf{Indicator} & \textbf{Investment} \\ \hline \text{US GDP} & \text{US stocks} \\ \hline \text{Non-Farm Payrolls} & \text{US stocks} \\ \hline \text{Non-Farm Payrolls} & \text{US stocks} \\ \hline \text{Initial Jobless Claims} & \text{US stocks} \\ \hline \text{Industrial Production Index} & \text{Industrial stocks} \\ \hline \text{Durable Goods Orders} & \text{Industrial stocks} \\ \hline \text{Total Retail Sales} & \text{Consumer stocks} \\ \hline \text{Michigan CSI} & \text{Consumer stocks} \\ \hline \text{US Building Permits} & \text{Real Estate stocks} \\ \hline \end{array}$$

We have relied on our common sense intuition, and the invaluable advice of our experienced volunteer Community Trader, @whaffner. A more sophisticated approach might include doing a regression analysis to measure the strength of historic and ongoing correlations, but for now, that’s beyond the scope of this pilot project.

Since we only want to be exposed to market movements correlated with the economic indicator we are forecasting, and market prices may change due to a wide variety of other factors, we plan to hedge our investments.

For example, if we own US stocks as part of our US GDP trade, then we will hedge by shorting international stocks. The purpose of the hedge is to offset any movements in our US stock investment that are not related to the US GDP release. If US GDP is better than expected, then we expect US stocks to rise while International stocks won’t be directly affected. However, if a war started we expect the price of US stocks and international stocks to decline. In that case, we hope that our loss in US stocks is offset by our gains due to shorting international stocks.

The table below summarizes the long/short pairs that we aim to be exposed to for each indicator. Note that these pairs are the product of fairly coarse-grained analysis and have not yet undergone a strict, rigorous analysis, but they should provide a clear enough signal to suit our current purpose.

$$\begin{array}{||c|cc|cc||} \hline & \textbf{If Metaculus} \gt \rlap{\textbf{Consensus}} && \textbf{If Metaculus} \lt \rlap{\textbf{Consensus}} \\ \hline \textbf{Indicator} & \textbf{Long} & \textbf{Short} & \textbf{Long} & \textbf{Short} \\ \hline \textbf{US GDP} & \text{US stocks} & \text{International stocks} & \text{International stocks} & \text{US stocks} \\ \hline \textbf{Non-Farm Payrolls} & \text{US stocks} & \text{International stocks} & \text{International stocks} & \text{US stocks} \\ \hline \textbf{Initial Jobless Claims} & \text{International stocks} & \text{US stocks} & \text{US stocks} & \text{International stocks} \\ \hline \textbf{Industrial Production Index} & \text{Industrials} & \text{US stocks} & \text{US stocks} & \text{Industrials} \\ \hline \textbf{Durable Goods Orders} & \text{Industrials} & \text{US stocks} & \text{US stocks} & \text{Industrials} \\ \hline \textbf{Total Retail Sales} & \text{Consumer} & \text{US stocks} & \text{US stocks} & \text{Consumer} \\ \hline \textbf{Michigan CSI} & \text{Consumer} & \text{US stocks} & \text{US stocks} & \text{Consumer} \\ \hline \textbf{US Building Permits} & \text{Real Estate} & \text{US stocks} & \text{US stocks} & \text{Real Estate} \\ \hline \end{array}$$

Trading Plan Implementation

To implement our strategy, we will use exchange traded funds (ETFs). The table below summarizes the ETFs we plan to use to go long.

$$\begin{array}{||c|c|c||} \hline \hline \textbf{Market} & \textbf{Long ETF} & \textbf{Name} \\ \hline \text{US stocks} & \text{SPY} & \text{iShares core S&P 500 ETF} \\ \hline \text{International Stocks} & \text{IXUS} & \text{iShares core MSCI Total International stock} \\ \hline \text{Industrials} & \text{XLI} & \text{Industrial Select SPDR fund} \\ \hline \text{Consumer Discretionary} & \text{XLY} & \text{Consumer discretionary select SPDR fund} \\ \hline \text{US Home Construction} & \text{ITB} & \text{iShares US home construction} \\ \hline \end{array}$$

Getting short exposure is slightly more complicated since the dedicated account for this project does not enable shorting directly. Instead, we will buy ETFs that short different sectors. Complicating matters slightly, some of these short ETFs use leverage, which we show in the table below.

$$\begin{array}{||c|c|c|c||} \hline \hline \textbf{Market} & \textbf{Short ETF} & \textbf{Leverage} & \textbf{Name} \\ \hline \text{US stocks} & \text{SH} & \text{-1x} & \text{ProShares Short S&P 500} \\ \hline \text{International Stocks} & \text{EFU} & \text{-2x} & \text{ProShares UltraShort MSCI EAFE} \\ \hline \text{Industrials} & \text{SIJ} & \text{-2x} & \text{ProShares UltraShort industrials} \\ \hline \text{Consumer Discretionary} & \text{SZK} & \text{-2x} & \text{ProShares UltraShort Consumer Goods} \\ \hline \text{US Home Construction} & \text{REK} & \text{-1x} & \text{ProShares Short Real Estate} \\ \hline \end{array}$$

Because of the leverage in some of the short ETFs above, we will scale our longs/shorts so that the exposure is in a 1:1 ratio. The table below summarizes the relevant sizing of the trades.

$$\begin{array}{||c|cc|cc||} \hline & \textbf{If Metaculus} \gt \rlap{\textbf{Consensus}} && \textbf{If Metaculus} \lt \rlap{\textbf{Consensus}} \\ \hline \textbf{Indicator} & \textbf{Long} & \textbf{Short} & \textbf{Long} & \textbf{Short} \\ \hline \textbf{US GDP} & \text{2x SPY} & \text{1x EFU} & \text{1x IXUS} & \text{1x SH} \\ \hline \textbf{Non-Farm Payrolls} & \text{2x SPY} & \text{1x EFU} & \text{1x IXUS} & \text{1x SH} \\ \hline \textbf{Initial Jobless Claims} & \text{1x IXUS} & \text{1x SH} & \text{2x SPY} & \text{1x EFU} \\ \hline \textbf{Industrial Production Index} & \text{1x XLI} & \text{1x SH} & \text{2x SPY} & \text{1x SIJ} \\ \hline \textbf{Durable Goods Orders} & \text{1x XLI} & \text{1x SH} & \text{2x SPY} & \text{1x SIJ} \\ \hline \textbf{Total Retail Sales} & \text{1x XLY} & \text{1x SH} & \text{2x SPY} & \text{1x SZK} \\ \hline \textbf{Michigan CSI} & \text{1x XLY} & \text{1x SH} & \text{2x SPY} & \text{1x SZK} \\ \hline \textbf{US Building Permits} & \text{1x ITB} & \text{1x SH} & \text{1x SPY} & \text{1x REK} \\ \hline \end{array}$$

As an example, if the Metaculus Prediction for U.S. GDP is above the consensus and we have $1500 in our account, then we will aim to buy $1000 of SPY and $500 of EFU. Our SPY position will be 2x bigger than the EFU position in order to match the exposures. Said another way, the $500 EFU position provides (-2x) x $500 = -$1000 exposure to international stocks hedging our $1000 SPY exposure.

Size

In theory, we could scale the size of our bets in proportion to our confidence that we will make money. But for the time being, we plan to adopt a simple approach of investing the full amount of our account – the entirety of the Trade Signal prize pool at any given moment – for every trade. If there is no difference between our forecast and the consensus forecast, then we won’t make a trade. Our source for the consensus forecast and the relevant significant digits will be Marketwatch.

Splits

When we can make two or more trades at the same time, then we’ll split our investment evenly between the options. If any leg of one trade cancels a leg of another trade, then we won’t trade that leg and will scale the remaining leg(s) appropriately.

Timing

In an ideal world, we would initiate our trade just before an economic release and unwind the trade immediately afterwards. In practice, many releases occur in the morning before US stock exchanges open, so we generally plan to initiate trades 30 minutes before the market closes the day before a release and to unwind the trade within the first 30 minutes of trading on the day of the release. If a release is scheduled during the trading day, then we will initiate and close the trade 30 minutes before and after the release.

Trading Mechanisms

We will use market orders to make trades. The market liquidity for all of the above ETFs is significantly larger than our order size, so we don’t expect to have a market impact. With larger sums, we would have to rethink this approach.

Some of the indicators in the Trade Signal Tournament, like the EIA forecast, are not expected to be market moving news, so we will not trade on them. Also, initial jobless claims releases occur once a week, but the corresponding Metaculus questions ask about the 4 week average. Due to the mismatch, we currently do not plan to trade on this release.

We will trade based on the Metaculus Prediction, as opposed to the community median.

Frictional Costs

Minimizing frictional costs is important for the success of any trading strategy. While our brokerage account does not charge commissions, there are other frictional costs.

All publicly traded investments have a bid/ask spread, which is the difference between the price to buy or sell the investment. The long ETFs we will use are quite liquid with spreads of around 0.01% of the price. However, the short ETFs have spreads of approximately 0.2% to 0.5%, which is substantial.

We believe another dominant frictional cost is likely to be associated with the leveraged short ETFs that we plan to use. We expect these ETFs to lose their value over any substantial holding period in part due to the cost of carry having to do with rolling futures contracts. For instance, we note that VXX has lost approximately 85% of its value over the last 5 years while the VIX index, that it provides short exposure to, is about the same today as it was 5 years ago.

Schedule

Many of the questions were originally scheduled to close a week or two before the economic release. We’ve rescheduled the closing dates for questions that will be used to trade so that they remain open right up until the trade is placed. Needless to say, this is important to ensure that the forecasts incorporate all the latest available information.

The table below shows the 14 releases that we currently plan to trade on. All of the corresponding questions are already live and cover economic indicators through September 30, but they may not be resolved until the end of October.

$$\begin{array}{||c|c|c||} \hline \hline \textbf{Indicator} & \textbf{Period covered} & \textbf{Resolution date} \\ \hline \text{US GDP} & \text{Q3} & \text{10-28-21} \\ \hline \text{Non-Farm Payrolls} & \text{Aug} & \text{9-3-21} \\ \hline \text{Non-Farm Payrolls} & \text{Sept} & \text{10-8-21} \\ \hline \text{Industrial Production Index} & \text{Aug} & \text{9-15-21} \\ \hline \text{Industrial Production Index} & \text{Sept} & \text{10-18-21} \\ \hline \text{Durable Goods Orders} & \text{July} & \text{8-25-21} \\ \hline \text{Durable Goods Orders} & \text{Aug} & \text{9-27-21} \\ \hline \text{Durable Goods Orders} & \text{Sept} & \text{10-27-21} \\ \hline \text{Total Retail Sales} & \text{Aug} & \text{9-16-21} \\ \hline \text{Total Retail Sales} & \text{Sept} & \text{10-15-21} \\ \hline \text{Michigan CSI} & \text{Aug} & \text{9-17-21} \\ \hline \text{Michigan CSI} & \text{Sept} & \text{10-15-21} \\ \hline \text{US Building Permits} & \text{Aug} & \text{9-21-21} \\ \hline \text{US Building Permits} & \text{Sept} & \text{10-19-21} \\ \hline \end{array}$$

Next Steps

We appreciate the efforts of all our participating forecasters in the Trade Signal Tournament thus far, and we look forward to reporting trade results over the coming weeks.

This document is for informational purposes only and is not investment advice.