Crash insurance: shorting the index after a close below EMA 100

Idea in brief

Laurens Bensdorp trades 55 strategies on stocks and ETFs: trend-following and mean reversion, long and short. In a crash, long systems lose at the same time. For this Bensdorp has three kinds of hedges with different purposes: long volatility on VXX for a sudden panic, mean reversion shorts for a prolonged bear market, and a simple trend-following index short for a catastrophe. This card is about the last one.

The rule from #042: when the index's daily close drops below the 100-day EMA, the strategy sells with a stop order and then manages the position with a trailing stop. You should trade an index future or another proxy for the whole market, not individual stocks. In #025 the author explains how to choose the speed of such a hedge and names other windows: 20, 30–50 and 200 days.

The strategy loses money on its own. It is judged as insurance: whether it pays when the worst happens, and how much it costs in calm years.

Why it might work

According to the author, the ideal trend-following short trade makes more the longer you stay in it. Such trades are possible in prolonged bear markets like 1929–1932, and that is exactly when a long portfolio loses the most.

Bensdorp calls this a strategy with a purpose. A hedge must make money when the market falls and is allowed to lose when the market rises. Bensdorp compares it to health insurance: the premium is paid every month, and the payout comes when you need surgery. A hedge with a short window enters earlier, but its premium, according to the author, is noticeably higher too: more false sells in a rising market.

The rule has a weakness, and the author mentions it. When the market falls 15% in three or four days, the trend hedge is still waiting for the average to be crossed and has no position. For this case the author has a separate hedge on VXX. The interview contains no test results for the trend-following short itself, only estimates of drawdown and return over a long history.

Rules

Instrument and signal (author, Finetiq stop level)

// daily bars of the E-mini S&P 500 future                                // author
// or another proxy for the whole market: Russell 2000 or 3000, S&P 500, Dow   // author
// not stocks: in a long-held short, company risk and dividend payments grow   // author

EMA100 = EMA(Close, 100)                  // author (#042)
IF Close > EMA100 THEN Armed = TRUE       // Finetiq: one attempt per drop below the average

IF position = FLAT AND Armed AND Close < EMA100
    SigLow = Low of the first bar that closed below EMA100
    SELL SHORT STOP at SigLow             // author: "sells on a stop"; Finetiq: stop level
    // Finetiq, reading B: SELL SHORT AT NEXT BAR OPEN
IF Close > EMA100 THEN CANCEL
// after entry Armed = FALSE until the next close above EMA100

Managing the position (author: trail; Finetiq parameters)

// author (#042): from there a trailing stop manages the position, type and distance not named
// author (#025): stay in the position until there are signs that most of the decline is over

TrailStop = Highest(High, 20)[1]              // Finetiq: starting window of 20 days
TrailStop = MIN(TrailStop, TrailStop[1])      // Finetiq: the stop only moves down
EXIT STOP at TrailStop
// the author names no separate stop-loss. Finetiq: the first trail level is the initial stop

// Finetiq, exit variants for testing
// B: IF Close > EMA100 THEN EXIT AT NEXT BAR OPEN
// C: TrailStop = lowest Low since entry + 3 * ATR(20)

Variant B. Hedge speed per #025 (author, Finetiq parameters)

// author: first decide when protection is needed
// immediately as the decline starts: a short window of about 20 days
// only in a long downtrend: about 200 days
// "fairly fast": a Keltner channel or a simple average over 30, 40 or 50 days

Fast = Close < SMA(Close, 50)     // author: 30–50 days; Finetiq: 50 and SMA instead of the channel
Slow = Close < EMA(Close, 200)    // author: 200 days (in #042 "100 or 200"); Finetiq: EMA
// entry and trail as in the main rule

// author: do not add conditions such as RSI above 50. An extra variable may
// keep you out of the position when the hedge is needed most

Size and evaluation (author, Finetiq numbers)

// author: hedges run alongside the long strategies, and the author gives them a larger share than is customary
// the author does not state the size of the share
HedgeNotional = h * LongBookNotional              // Finetiq: h = 0.3, test 0.2 and 0.5
Contracts     = HedgeNotional / (Price * PointValue)

// evaluation (author): does it make money in a bear market and how much does it give back in a bull market
// the strategy's own CAGR and maximum drawdown are not the criterion
CrisisWindows = 2000-09..2002-10, 2007-10..2009-03, 2020-02..2020-04, 2022-01..2022-10   // Finetiq
Report = hedge P&L in each window, P&L in years when the index rose,
         portfolio drawdown with and without the hedge

Parameters

What to test

1. Crisis windows. Calculate the hedge's result in 2000–2002, 2008, February–April 2020 and 2022 next to the long portfolio's loss over the same dates. Separately, calculate how much the hedge lost in years when the index rose: that is the price of the insurance.
2. Portfolio with and without the hedge. Take a long portfolio (buy and hold of the S&P 500 or your own breakout strategy) and add the hedge at a share of 0.2, 0.3, 0.5. Compare the maximum drawdown, time under water and return. This is the main test, and a standalone hedge equity curve does not replace it.
3. Stop entry or open entry. Reading A (a stop below the low of the signal bar) versus B (the next day's open). If price chops around the average, A should filter out some of the false sells.
4. Which trail. A 20-day high versus a 3 ATR chandelier and an exit on a close above EMA 100. Look at what share of the profit in crisis windows is kept and how many times the hedge re-enters.
5. Average neighborhood. EMA 80, 100, 150, 200 and the fast 30, 40, 50 days from #025. For each length compare earnings in crises and losses in calm years. A hedge is chosen by its purpose, not by maximum return.
6. V-shaped crash. For 2020 find the entry date, the exit date and the result. The author says that in the first days of a panic the trend hedge is not in the market. If the entry landed at the bottom, the hedge also lost on the rebound.
7. Cost of holding. The short is held for months. For a future account for rolls, for a CFD the swap and dividend adjustments for every night.

Platform notes

TradingView (Pine Script)

• A strategy trades one symbol. The hedge and the long portfolio cannot be combined in one backtest: export the trades of both and add the equity curves outside the platform.
• strategy.entry("H", strategy.short, stop = sigLow) stays active until canceled. When the close is back above EMA 100, remove the order via strategy.cancel("H"), otherwise it will fill later, with no signal behind it.
• EMA 100 needs a warm-up of 300–400 daily bars. To test on 2000 and 2008 you need continuous futures history from the late 1990s, and its depth depends on the plan.
• A run on the S&P 500 cash index is possible, but the index itself cannot be traded. Results on the index and on the future will differ by rolls and basis.

MultiCharts and TradeStation (EasyLanguage)

• Sell Short next bar at SigLow stop lives for one bar. The order must be sent every day while the close is below EMA 100 and there is no position.
• Back-adjusting a future by subtraction shifts prices by a constant. The crossover with the average does not suffer from this, but a position size as a percentage of price is distorted on old history.
• To see the hedge together with the long strategies you need a portfolio module: Portfolio Trader in MultiCharts, Portfolio Maestro in TradeStation.
• A futures daily bar in TradeStation closes at settlement, a 1440-minute bar at the last trade. A close near EMA 100 on the two bar types sometimes gives a signal on different days.

MetaTrader 5 (MQL5)

• On a netting account a short on the same symbol as a long position simply reduces it. The hedge must be run on a separate symbol or on a hedging account.
• An index CFD is not a future. At many brokers short positions in index CFDs are charged dividend adjustments and swap. The author's argument that an index spares you dividend payments on a short does not hold here.
• A broker's daily CFD history often starts after 2008. Testing on crises will have to be done on external data.
• Take the EMA from the closed bar (index 1) and check the signal once when a new daily bar appears. Short Sunday bars at some brokers shift EMA 100.

Where the idea can break

• The author did not describe the trailing stop. The hedge's result in a crisis depends heavily on this choice, and the parameters in the card are ours.
• The author's figures are qualitative or approximate. There is no test report, number of trades or result by window. The two episodes give different history start dates and drawdowns: since 1920 and 99% in #042, since 1928 and 98% in #025.
• A test on 1929–1932 is only possible on the cash index: there were no futures then, and shorting conditions and costs were different. In #025 the author says that mean reversion systems cannot be tested on 1929 and 1987 because of different commissions and tick sizes. The author does not apply this caveat to the hedge.
• In a V-shaped crash the signal comes late. After a fast rebound the hedge is closed by the trail at a loss, and the insurance did not pay.
• The insurance premium in a long bull market is paid for years in a row. It is easy to give up on the hedge in the third losing year. The author is against switching strategies off based on recent results, but that is the author's position, not a tested rule.
• A hedge on the S&P 500 protects a stock portfolio only as far as the portfolio resembles the index. A portfolio of small caps or a single sector can fall harder.
• The author does not state the share of hedges in their own portfolio. In both interviews the author promotes their courses and books.