Backtesting Delisted Equities Without Look-Ahead Slippage
Summary
The document considers transaction-cost assumptions for daily, medium- to long-term equity backtests, including trades in companies that are later delisted. The question describes using an estimated bid–ask spread and calibrating exit slippage against actual trades, with different assumptions for ordinary exits, filing days, and stop-losses. It asks whether a special slippage adjustment should apply near delisting, when calibration data may be unavailable.
The response cautions that assigning special slippage because a security is known to have delisted can introduce look-ahead bias. It suggests handling the position using observable post-delisting outcomes: follow the security if it continues trading over the counter, or mark it to zero if it is halted. The document offers no empirical comparison of cost models or calibration procedure for delisting events. Its practical lesson is to avoid using future delisting knowledge to alter historical execution costs and to represent the security’s actual subsequent trading or halt outcome.
Key ideas
- The question models transaction costs using an estimated spread and calibrated exit slippage.
- Applying special slippage based on eventual delisting can create look-ahead bias.
- A delisted security may continue to be valued using its over-the-counter price.
- If trading is halted, the response suggests marking the position value to zero.
Tags
Full text
# Realistic slippage estimation for delisted equities
# Realistic slippage estimation for delisted equities
I use various medium- to long-term strategies, which are primarily based on fundamental indicators and daily eod market prices. For backtesting I'm using the following model to simulate transaction costs:
$buy_a = close + ba_{\text{spread}}$
$sell_a = (close - ba_{\text{spread}})\cdot slippage \; (\text{exit reason})$
Where $buy_a$ and $sell_a$ are the actual prices I'm using in my backtest, $close$ is the adjusted close-mid price that I'm getting from my market data providers, and $ba_{\text{spread}}$ is an model-based estimator for the bid–ask spread. For slippage, I use a different fraction depending on why the trade was closed. For signal-triggered exits on “normal” days, I use 1; for signal-triggered exits on SEC filing days (as there is usually much higher volatility) or if a stop-loss was triggered (since you will rarely get the actual stop-loss limit), I use a ratio < 1, which I regularly calibrate based on real trades. This means I compare the actual selling price I achieved with the closing price from my providers and calculate the average of:
$\frac{sell_a}{close - ba_{\text{spread}}}$
The first question is whether this model is generally appropriate and delivers realistic results. If so, I also wanted to ask how delisted companies can be considered in this context. Basically, I could extend the same procedure to delistings, but of course I have never had a real trade in which a stock was delisted while I held it, so I can't calibrate anything and would have to guess. I am also unsure when this “delisting slippage" should be applied. The options would be for all delisted tickers, for all delisted tickers whose exit date is exactly on the day of the official delisting from public exchanges, or even already a few days before (because the impending delisting is likely to have already been priced in beforehand leading to higher volatility).
## Answer by pyCthon (score 1)
https://quant.stackexchange.com/a/85364
Why would the slippage be different then any other stock at the time it was listed? Applying a special “delisting slippage" is essentially adding look ahead bias.
When a stock becomes delisted you can do one of two things, track the OTC price if it's still being traded OTC or mark the value down to zero if it's halted.Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.